prompt
stringlengths 16
15.6k
| completion
stringlengths 4
6
|
|---|---|
Title: Hello. I'm 17, I've loved computers since I was 5. I'm a pretty decent programmer, I've done some big projects, worked with teams on extended projects, and have a pretty decent understanding of programming/linux/networking/security/etc. I generally write C, Rust, Python, and TS. I attend a STEM school that specialises in CS and engineering, and I'm in my first year of A Levels (UK) studying Maths, Further Maths, Physics, and Computer Science. I'm on track to get As and A*s.<p>The plan 6 months ago was - get good grades, go to a good uni, get a CS degree while doing programming to pay for uni/do a degree apprenticeship, and go into a CS career. Without LLMs messing the world up, I think it's a pretty good plan. I'm a good programmer for my age and I'm academic.<p>However, with recent developments, that seems borderline foolish. I'm interested in physics and engineering, I think I could enjoy a career in either, and I've still got time to pivot to either. I might be able to write better code than GPT-10 - might - but there would surely be increased competition. I also just don't find asking a LLM in the "correct" way and having it spit out code trained on millions of other programmers interesting in the slightest. I worry that actually knowing how to write code without a LLM will go the way of knowing assembly and having intimate knowledge of a computers internals.<p>I'm frankly tired of hearing about LLMs, but I know that I can't just pretend they don't exist. I have less than a year to make a decision on which uni courses I apply to and I don't know whether I should abandon a sinking ship or hope it's alright and it stops here, as an aid but not something which ruins programming.<p>I've seen people saying "But we'll still need to translate client requirements to prompts" - 1. I doubt that'll pay well, and while that's not my primary motivation it is something to consider before committing myself, 2. I'm not interested in being a human-bullshit to AI-bullshit translator.<p>And yes, there's more to CS than programming, but not in a way that pays the bills. If I pivoted I'd still write code "The old fashioned way" for fun.<p>So: What do I do? I'd appreciate some advice.<p>Thanks :)
Upvote: | 43 |
Title: There are a lot of Python to web app frameworks going around these days but I wanted something that was a little more lightweight that just generates HTML pages and can be embedded in Flask or other Python web servers incrementally.<p>PyVibe uses Python components to construct a page with styling that you can use in Flask, in a static site, or even in Pyodide.
Upvote: | 229 |
Title: I've been playing around with <a href="https://github.com/zphang/minimal-llama/">https://github.com/zphang/minimal-llama/</a> and <a href="https://github.com/tloen/alpaca-lora/blob/main/finetune.py">https://github.com/tloen/alpaca-lora/blob/main/finetune.py</a>, and wanted to create a simple UI where you can just paste text, tweak the parameters, and finetune the model quickly using a modern GPU.<p>To prepare the data, simply separate your text with two blank lines.<p>There's an inference tab, so you can test how the tuned model behaves.<p>This is my first foray into the world of LLM finetuning, Python, Torch, Transformers, LoRA, PEFT, and Gradio.<p>Enjoy!
Upvote: | 449 |
Title: ChatLLaMA is an experimental chatbot interface for interacting with variants of Facebook's LLaMA. Currently, we support the 7 billion parameter variant that was fine-tuned on the Alpaca dataset. This early versions isn't as conversational as we'd like, but over the next week or so, we're planning on adding support for the 30 billion parameter variant, another variant fine-tuned on LAION's OpenAssistant dataset and more as we explore what this model is capable of.<p>If you want deploy your own instance is the model powering the chatbot and build something similar we've open sourced the Truss here: <a href="https://github.com/basetenlabs/alpaca-7b-truss">https://github.com/basetenlabs/alpaca-7b-truss</a><p>We'd love to hear any feedback you have. You can reach me on Twitter @aaronrelph or Abu (the engineer behind this) @aqaderb.<p>Disclaimer: We both work at Baseten. This was a weekend project. Not trying to shill anything; just want to build and share cool stuff.
Upvote: | 402 |
Title: Hey! I wanted to research the possibilities of participating in open-source projects, preferably related to audio or sound processing. Do you have any projects to recommend? Preferably ones that are either technically interesting, solve a useful problem or have lively community and are in active development.<p>I'm not specifying anything more (language, etc.), so that I can get as wide of a selection, as possible and so that this post is useful for others more.
Upvote: | 54 |
Title: So lately I've been dabbling in a lot of stuff that requires a lot of CPU compute, 0 GPU and can be basically linearly scaled across threads and nodes in a cluster.<p>Now using only my own box is proving to be a bottleneck so I've been thinking of either using AWS spot instances or building my own mini-cluster (2-3 machines + switch) at home. Does it make sense to go cloud (even spot) when I would aim at high utilization?<p>As for the potential node spec:<p>- Ryzen 4500/5500 (seems best perf/$)<p>- Some mATX AM4 mobo with integrated GPU<p>- 2x8GB RAM<p>- mATX case, the smaller the better. ITX seems pricier.<p>All the box does is basically run the k8s pod(s).<p>WDYT?
Upvote: | 42 |
Title: I am thinking of trying out Hetzner for hosting front-ends, back-ends. I have some questions about the workflow on Hetzner.<p>How do you<p>- deploy from source repo? Terraform?<p>- keep software up to date? ex: Postgres, OS<p>- do load balancing? built-in load balancer?<p>- handle scaling? Terraform?<p>- automate backups? ex: databases, storage. Do you use provided backups and snapshots?<p>- maintain security? built-in firewall and DDoS protection?<p>If there is any open source automation scripts please share.
Upvote: | 113 |
Title: Hey HN - we're Daniel, Taner, and Nic, and we're building Flower (<a href="https://flower.dev/">https://flower.dev/</a>), an open-source framework for training AI on distributed data. We move the model to the data instead of moving the data to the model. This enables regulatory compliance (e.g. HIPAA) and ML use cases that are otherwise impossible. Our GitHub is at <a href="https://github.com/adap/flower">https://github.com/adap/flower</a>, and we have a tutorial here: <a href="https://flower.dev/docs/tutorial/Flower-0-What-is-FL.html">https://flower.dev/docs/tutorial/Flower-0-What-is-FL.html</a>.<p>Flower lets you train ML models on data that is distributed across many user devices or “silos” (separate data sources) without having to move the data. This approach is called federated learning.<p>A silo can be anything from a single user device to the data of an entire organization. For example, your smartphone keyboard suggestions and auto-corrections can be driven by a personalized ML model learned from your own private keyboard data, as well as data from other smartphone users, without the data being transferred from anyone’s device.<p>Most of the famous AI breakthroughs—from ChatGPT and Google Translate to DALL·E and Stable Diffusion—were trained with public data from the web. When the data is all public, you can collect it in a central place for training. This “move the data to the computation” approach fails when the data is sensitive or distributed across organizational silos and user devices.<p>Many important use cases are affected by this limitation:<p>* Generative AI: Many scenarios require sensitive data that users or organizations are reluctant to upload to the cloud. For example, users might want to put themselves and friends into AI-generated images, but they don't want to upload and share all their photos.<p>* Healthcare: We could potentially train cancer detection models better than any doctor, but no single organization has enough data.<p>* Finance: Preventing financial fraud is hard because individual banks are subject to data regulations, and in isolation, they don't have enough fraud cases to train good models.<p>* Automotive: Autonomous driving would be awesome, but individual car makers struggle to gather the data to cover the long tail of possible edge cases.<p>* Personal computing: Users don't want certain kinds of data to be stored in the cloud, hence the recent success of privacy-enhancing alternatives like the Signal messenger or the Brave browser. Federated methods open the door to using sensitive data from personal devices while maintaining user privacy.<p>* Foundation models: These get better with more data, and more diverse data, to train them on. But again, most data is sensitive and thus can't be incorporated, even though these models continue to grow bigger and need more information.<p>Each of us has worked on ML projects in various settings, (e.g., corporate environments, open-source projects, research labs). We’ve worked on AI use cases for companies like Samsung, Microsoft, Porsche, and Mercedes-Benz. One of our biggest challenges was getting the data to train AI while being compliant with regulations or company policies. Sometimes this was due to legal or organizational restrictions; other times, it was difficulties in physically moving large quantities of data or natural concerns over user privacy. We realized issues of this kind were making it too difficult for many ML projects to get off the ground, especially in domains like healthcare and finance.<p>Federated learning offers an alternative — it doesn't require moving data in order to train models on it, and so has the potential to overcome many barriers for ML projects.<p>In early 2020, we began developing the open-source Flower framework to simplify federated learning and make it user-friendly. Last year, we experienced a surge in Flower's adoption among industry users, which led us to apply to YC. In the past, we funded our work through consulting projects, but looking ahead, we’re going to offer a managed version for enterprises and charge per deployment or federation. At the same time, we’ll continue to run Flower as an open-source project that everyone can continue to use and contribute to.<p>Federated learning can train AI models on distributed and sensitive data by moving the training to the data. The learning process collects whatever it can, and the data stays where it is. Because the data never moves, we can train AI on sensitive data spread across organizational silos or user devices to improve models with data that could never be leveraged until now.<p>Here’s how it works: (0) Initialize the global model parameters on the server; (1) Send the model parameters to a number of organizations/devices (client nodes); (2) Train model locally on the data of each organization/device (client node); (3) Return the updated model parameters back to the server; (4) On the server, aggregate the model updates (e.g., by averaging them) into a new global model; (5): Repeat steps 1 to 4 until the model converges.<p>This, of course, is more challenging than centralized learning: we must move AI models to data silos or user devices, train locally, send updated models back, aggregate them, and repeat. Flower provides the open-source infrastructure to easily do this, as well as supporting other privacy-enhancing technologies (PETs). It is compatible with PyTorch, TensorFlow, JAX, Hugging Face, Fastai, Weights & Biases and all the other tools used in ML projects regularly. The only dependency on the server side is NumPy, but even that can be dropped if necessary. Flower uses gRPC under the hood, so a basic client can easily be auto-generated, even for most languages that are not supported today.<p>Flower is open-source (Apache 2.0 license) and can be run in all kinds of environments: on a personal workstation for development and simulation, on Google Colab, on a compute cluster for large-scale simulations or on a cluster of Raspberry Pi’s (or similar devices) to build research systems, or deployed on public cloud instances (AWS, Azure, GCP, others) or private on-prem hardware. We are happy to help users when deploying Flower systems and will soon make this even easier through our managed cloud service.<p>You can find PyTorch example code here: <a href="https://flower.dev#examples">https://flower.dev#examples</a>, and more at <a href="https://github.com/adap/flower/tree/main/examples">https://github.com/adap/flower/tree/main/examples</a>.<p>We believe that AI technology must evolve to be more collaborative, open and distributed than it is today (<a href="https://flower.dev/blog/2023-03-08-flower-labs/">https://flower.dev/blog/2023-03-08-flower-labs/</a>). We’re eager to hear your feedback, experiences regarding difficulties in training, data access, data regulation, privacy and anything else related to federated (or related) learning methods!
Upvote: | 180 |
Title: Hey HackerNews,<p>Today I'd like to share my open source project, Moonshine!<p>Pretrained vision models are a popular way to reduce how much data you need and to speed up training, but for remote sensing (i.e. aerial, satellite) it can be a challenge to find good weights. Why use Moonshine?<p>1. Pretrained on multispectral data: Existing popular CV packages are nice to use but are trained on ImageNet or similar, meaning that not only is the domain of the pretraining different, but you may be restricted to only 3 channels. For remote sensing platforms with multispectral data, this might be a non-starter. Moonshine includes models specifically trained for your remote sensing problems.<p>2. Focus on usability: There are academic releases of specialized models that do support remote sensing data, but often they are difficult to use. They might be hard to install, and often are supported by a grad student who is more focused on publication than software. A core tenant of Moonshine is that it should be easy to use.<p>I've been working on this project for nearly a year now and I'm really excited to show it off and most importantly get feedback! I have big plans for what to build in the future, but this set of features was the smallest one I could think of that would provide some value.<p>Docs: <a href="https://moonshineai.readthedocs.io/en/latest/index.html" rel="nofollow">https://moonshineai.readthedocs.io/en/latest/index.html</a>
Github: <a href="https://github.com/moonshinelabs-ai/moonshine">https://github.com/moonshinelabs-ai/moonshine</a>
Upvote: | 86 |
Title: Hey HN! We launched Zapier way back in 2012 on HN: <a href="https://news.ycombinator.com/item?id=4138415" rel="nofollow">https://news.ycombinator.com/item?id=4138415</a> and thought we'd return home to announce something special and hopefully exciting :) We are trying to finally live up to the "API" in our name with Zapier's first universal API:<p>Natural Language Actions – <a href="https://zapier.com/l/natural-language-actions" rel="nofollow">https://zapier.com/l/natural-language-actions</a><p>API docs – <a href="https://nla.zapier.com/api/v1/docs" rel="nofollow">https://nla.zapier.com/api/v1/docs</a><p>(to be fair, we have published APIs before that can access Zapier data, but never before one devs can use to directly call the 5k+ apps / 20k+ actions on our platform)<p>For example, you can use the API to:<p><pre><code> * Send messages in Slack
* Retrieve a row in a Google Sheet
* Draft a reply in Gmail
* ... and thousands more actions with one universal API
</code></pre>
We optimized NLA for use cases that receive user input in natural language (think chatbots, assistants, or any product/feature using LLMs) -- but not strictly required!<p>Folks have asked for an API for 10 years and I've always been slightly embarrassed we didn't have one. We hesitated because we did not want to pass along our universe of complexity to end devs. With the help of LLMs we found some cool patterns to deliver the API we always wanted.<p>My co-founder/CTO Bryan did an interview with Garry on YC blog with more details: <a href="https://www.ycombinator.com/blog/building-apis-for-ai-an-interview-with-zapiers-bryan-helmig" rel="nofollow">https://www.ycombinator.com/blog/building-apis-for-ai-an-int...</a><p>We also published a LangChain integration to show off some possibilities:<p><pre><code> * Demo: https://www.youtube.com/watch?v=EEK_9wLYEHU
* Jupyter notebook: https://github.com/hwchase17/langchain/blob/master/docs/modules/utils/examples/zapier.ipynb
</code></pre>
We know the API is not perfect but we're excited and eager for feedback to help shape it.
Upvote: | 235 |
Title: This is a D&D map making tool I've been working on for a while now, but I just added some new features to the beta that I think HN users might find neat. When building a world map, you can use "Dynamic Brushes" to draw organic looking terrain. This is achieved entirely with svg filters and javascript canvas, no fancy libraries or anything. This came with a pretty large rewrite of some of the underlying code, so I'm sure there's a number of bugs I haven't come across, but I'd love to hear your opinions on it!
Upvote: | 157 |
Title: I started to learn the piano 1.5 year ago, as an adult who never studied music before.<p>It is a slow, peaceful journey and there is no silver bullet: practicing is the key for improvement.<p>I have a weekly class for music theory, which I enjoy a lot. What I seem to be a slow learner at is reading notes fast (treble and bass clef), and remembering the theory and logic behind the tones.<p>I don't practice much music theory outside the class. I consider building paper flashcards because music paper is quite specific.<p>What is your experience?
What process or tool would you recommend to learn music theory?<p>PS: sorry if some words feel weird, music vocabulary is so different between French and English
Upvote: | 321 |
Title: Hello HN! We’re Leo and Mandeep, founders of Datapane (<a href="https://datapane.com" rel="nofollow">https://datapane.com</a>).<p>We're building a way to create reports, dashboards, and web apps from your existing data using Python. Think of it as a combination of React and htmx, specifically designed for the Python data stack.<p>Our GitHub is <a href="https://github.com/datapane/datapane">https://github.com/datapane/datapane</a> and you can try building a report or app in ~2 minutes on Codespaces: <a href="https://try.datapane.com" rel="nofollow">https://try.datapane.com</a><p>We started building Datapane at our previous start-up, where we struggled to deliver ML model results to clients. Much to our surprise, the data science took less time than repeatedly creating reports by copying and pasting plots into PowerPoint decks.<p>It seemed absurd that we had to switch to PowerPoint or legacy BI tools like Tableau to share, and our initial goal was to programmatically generate reports using the datasets and plots we had in Python. To enable this, we started hacking on a Python-based UI framework for constructing HTML views from data-centric blocks – like plots, data tables, and layout components.<p>You can export these to standalone HTML files, or host them as a web app on somewhere like GitHub Pages or Fly.io. We recently also added the ability to connect Python functions to forms and front-end events so you can build web apps which run backend code. We handle the entire network and RPC layer, so you only need to write plain Python functions that take parameters and return other blocks.<p>You can check out an example of the code to create a simple app: <a href="https://github.com/datapane/examples/blob/main/apps/iris-plotter/app.py">https://github.com/datapane/examples/blob/main/apps/iris-plo...</a><p>Datapane’s philosophy is pretty different from other products in the space.<p>We wanted to keep things simple, but avoid the footguns our users faced with frameworks like Streamlit, where the reactive/network-aware model was hard to move beyond an MVP or POC. For backend interactivity, we believe the original web got a lot right, and unlike reactive models which rely on websockets, Datapane is unashamedly request/response. This takes inspiration from HTTP and our own experiences with htmx, which offers an elegant way to add interactivity to HTML. Under the hood, we actually compile down to a (gasp!) XML-based hypermedia format, akin to HTML, but tailored specifically for constructing data UIs.<p>The result is that not every change in your app requires a server round trip, as much of it can be pre rendered and most interactivity happens on the client-side. In addition to improving performance, this also makes running in production become 10x simpler.<p>This separation between the view and backend compute also makes Datapane modular. If our app server isn’t a good fit for your use-case, serve Datapane views from the web-framework of your choice (we’ve been hacking on serving views from Django). Want to compute blocks from inside Airflow or generate them on a schedule or from a webhook? Computation can happen out of band of the UI. You can even build and host apps from inside of Jupyter, where you can preview blocks live and convert notebook cells to blocks in your view.<p>We currently offer a hosting platform on <a href="https://datapane.com" rel="nofollow">https://datapane.com</a> for sharing reports publicly (free) or with your team (paid), and will be adding serverless app hosting support to it in the next few weeks.<p>Our ultimate goal is to create an open-source toolkit for building data products across the entire stack – from reports, to dashboards, to full-stack apps – all using 100% Python. You can see a few we’ve built already in our gallery: <a href="https://datapane.com/gallery" rel="nofollow">https://datapane.com/gallery</a><p>We’d love to hear your feedback.<p>Thanks!
Upvote: | 55 |
Title: Hi HN - Justin, Scott, and Barak here. We're excited to introduce Helicone (<a href="https://www.helicone.ai">https://www.helicone.ai</a>) an open-source logging solution for OpenAi applications. Helicone's one-line integration logs the prompts, completions, latencies, and costs of your OpenAI requests. It currently works with GPT, and can be integrated with one line of code. There’s a demo at <a href="https://www.helicone.ai/video">https://www.helicone.ai/video</a>.<p>Helicone's core technology is a proxy that routes all your OpenAI requests through our edge-deployed Cloudflare Workers. These workers are incredibly reliable and cause no discernible latency impact in production environments. As a proxy, we offer more than just observability: we provide caching and prompt formatting, and we'll soon add user rate limiting and model provider back off to make sure your app is still up when OpenAI is down.<p>Our web application then provides insights into key metrics, such as which users are disproportionately driving costs and what is the token usage broken down by prompts. You can filter this data based on custom logic and export it to other destinations.<p>Getting started with Helicone is quick and easy, regardless of the OpenAI SDK you use. Our proxy-based solution does not require a third party package—simply change your request's base URL from <a href="https://api.openai.com/v1" rel="nofollow">https://api.openai.com/v1</a> to <a href="https://oai.hconeai.com/v1" rel="nofollow">https://oai.hconeai.com/v1</a>. Helicone can be integrated with LangChain, LLama Index, and all other OpenAI native libraries. (<a href="https://docs.helicone.ai/quickstart/integrate-in-one-line-of-code">https://docs.helicone.ai/quickstart/integrate-in-one-line-of...</a>)<p>We have exciting new features coming up, one of which is an API to log user feedback. For instance, if you're developing a tool like GitHub Copilot, you can log when a user accepted or rejected a suggestion. Helicone will then aggregate your result quality into metrics and make finetuning suggestions for when you can save costs or improve performance.<p>Before launching Helicone, we developed several projects with GPT-3, including airapbattle.com, tabletalk.ai, and dreamsubmarine.com. For each project, we used a beta version of Helicone which gave us instant visibility into user engagement and result quality issues. As we talked to more builders and companies, we realized they were spending too much time building in-house solutions like this and that existing analytics products were not tailored to inference endpoints like GPT-3.<p>Helicone is developed under the Common Clause V1.0 w/ Apache 2.0 license so that you can use Helicone within your own infrastructure. If you do not want to self-host, we provide a hosted solution with 1k requests free per month to try our product. If you exceed that we offer a paid subscription as well, and you can view our pricing at <a href="https://www.helicone.ai/pricing">https://www.helicone.ai/pricing</a>.<p>We're thrilled to introduce Helicone to the HackerNews community and would love to hear your thoughts, ideas, and experiences related to LLM logging and analytics. We're eager to engage in meaningful discussions, so please don't hesitate to share your insights and feedback with us!
Upvote: | 166 |
Title: I was wondering when this day would arrive. "I" was pretty much the only way I used Reddit on my phone. IMO the new layouts (both 'old' and 'new') are far inferior - lower information density, ads, and more. This likely marks the start of a significant reduction in my Reddit consumption.<p><i>edit</i> Seems like it was rolled out with this update: https://old.reddit.com/r/reddit/comments/11zso11/an_improved_web_experience/
Upvote: | 64 |
Title: The classic mobile Reddit site is dead
Upvote: | 110 |
Title: I've been working at a company for a long time. It's grown from a startup into a big corporate company. I've managed to find my way inside the company, creating unique roles for myself and working on smaller growth initiatives. I've enjoyed the environment less and less as time has gone by, but I still felt there were valuable things I could do inside the company, especially with my knowledge as an early employee. Plus, they paid me well, the culture was mostly good, etc.<p>I recently got laid of which initially felt like a slap in the face. Once I caught my breath and started looking at options at other companies, I've realized how much more excited I am about joining a new small company than I am about anything I could have done at my old company.<p>I tried really hard to make myself happy the best I could inside that company, but it feels like a fresh of breath air realizing that my potential opportunities are much broader than that one company. It almost feels like Stockholm syndrome or something.<p>Anyway, I know I'm privileged to get laid off and be able to easily find a new job, but I just wanted to share in case anyone else is feeling the same way I was: trapped inside a company that I liked less and less over time.
Upvote: | 55 |
Title: We need to push the notion that "closed-source LLMs are super dangerous, an existential risk to humanity".<p>Basically we need to equate "safety" in LLMs to mean "being open-source".<p>OpenAI keeps talking about "safety" as the most important goal. If we define it to mean "open-source" then they will be pushed into a corner.<p>We are at a critical time period that will literally decide the outcome of humanity.
Upvote: | 119 |
Title: As a part time digital nomad, I use AirBnB quite oftern. It just happened to me that something broke during our stay. Host submited a claim to AirBnB about that (which is Ok) and they included a few other supposedly broken things. Fortunately I took some pictures before departure. I submited those to AirBnB as part of the mediation process they have. So far so good.<p>And the result? A single person reviewed the case and concluded that I should pay for everything. I asked them to review it again as they probably missed that picture. The response was that I agree in T&C that their decision is final and they will not be reviewing again. I asked to escalate the issue but the answer (from the same person who reviewed it) was the same.<p>I understand they have to draw the line somewhere but a single person deciding whether you need to pay something without any way of re-reviewing is probably too much. So probably keep that in mind as it can ruin the budget!
Upvote: | 57 |
Title: With GPT so hot in the news right now, and seeing lots of impressive demos, I'm curious to know, how are you actively using GPT to be productive in your daily workflow? And what tools are you using in tandem with GPT to make it more effective? Have you written your own tools, or do you use it in tandem with third party tools?<p>I'd be particularly interested to hear how you use GPT to write or correct code beyond Copilot or asking ChatGPT about code in chat format.<p>But I'm also interested in hearing about useful prompts that you use to increase your productivity.
Upvote: | 629 |
Title: I feel like Google’s search result quality has been steadily declining since at least five years, and from what I’m reading both on HN and on other platforms, people feel the same.<p>Add to that Alphabet’s notorious love for constantly killing products, and I’m wondering how long they will be able to keep this going?<p>With ChatGPT and other AI tools being introduced, will the need and reliance on classic search engines drop quickly or will it take years?<p>Where do you see this going?
Upvote: | 77 |
Title: Two years ago we me and my partners got an acquisition from a startup we created in the previous 2 years, while having a full time job, my background is software engineering. Since then I couldn't stop thinking on my next "idea". Rationally I know there is nothing that kills creativity like trying to be creative, one of my reference on the topic is Why Greatness Cannot Be Planned: The Myth of the Objective https://www.goodreads.com/en/book/show/25670869. That being said, this entrepreneurial goal has polluted everything in my life and the way I see the world. I stopped to genuinely enjoy experience since everything became a potential problem to solve, I started to be less present during interaction with other people and I pretended to have less time for things that I like because of some phantomatic idea I had to test. I recently started meditating with Sam Harris's Waking up, which is really helping on bringing more awareness in my life, not solving my problems but at least acknowledge those are there. My goal is to understand if I still really want to be an entrepreneur or I'm just projecting expectations from the shadow of the past success, if the answer is yes I want to keep my exploration in a more sane journey. I don't believe I'm the only one experiencing what I described, did anybody deal with it / are you dealing with this? What was your experience did you find something that worked for you?
Upvote: | 40 |
Title: Meet HomeByte (<https://homebyte.com>) - your partner in the home buying quest. We aim to be a better way to browse and find homes than on Zillow / Redfin. We are available in 23 states today and are adding more every month.<p>We just launched Reimagine – an AI powered interior designer that allows users to experiment with different styles to transform any room. With this feature, one can visualize a living room in an industrial style or a kitchen in a rustic lodge style, among others. This makes it easy for anyone to explore the potential of a home they are considering buying.<p>To get started, you can:
- Search for homes in your favorite neighborhood or city here - <https://homebyte.com/c/search> and click on the Purple “Reimagine this home” button on the home card.
- On homebyte.com, specify the address for any home that’s listed for sale. This will load the listing page and you can now click on the Purple “Reimagine this home”.<p>Example: <https://homebyte.com/search/details/2954-23rd-st-san-francisco-ca-94110/29510733><p>Here are some room transformation examples (<https://homebyte.com/reimagine/examples>) by Reimagine, and here’s what other folks are doing with Reimagine - <https://homebyte.com/reimagine>.<p>If you’d like to do this for your own home - head to <https://homebyte.com/reimagine/new> and upload your photos and Reimagine away!<p>We’d love to hear what you think and how we can make it better - you can either leave a comment here or write to us at [email protected]. We’re excited about the new AI powered features we’re adding to HomeByte. Stay tuned and thank you for your support!
Upvote: | 55 |
Title: Just curious
Upvote: | 66 |
Title: Hey there!<p>I just released the first version of a project I’ve been working on solves a very specific problem that perhaps only I have. I welcome any and all feedback, even if you just want to drop in to say that this is a hot piece of garbage!
Upvote: | 563 |
Title: CNBC just deleted 5 pages showing the current and historical 5 year Credit Default Swaps for:<p>- JPMorgan, deleted URL: https://www.cnbc.com/quotes/JPMCD5<p>- Bank of America, deleted URL: https://www.cnbc.com/quotes/BACCD5<p>- PNC, deleted URL: https://www.cnbc.com/quotes/PNCCD5<p>- Truist Financial, deleted URL: https://www.cnbc.com/quotes/TFCCD5<p>- Wells Fargo, deleted URL: https://www.cnbc.com/quotes/WFCCD5<p>Not deleted: Goldman Sachs https://www.cnbc.com/quotes/GSCD5, Deutsche Bank https://www.cnbc.com/quotes/DBCD5 .<p>You can easily use archive.org to check for yourself example: https://web.archive.org/web/20221009203204/https://www.cnbc.com/quotes/BACCD5<p>This might not seem like a big thing, but it is! This was one of the very few sources were you could see current credit default swaps data online. I would love to get an explanation from CNBC.<p>EDIT: I was actively crawling the pages in question, it did work until at least Friday, 24 March 2023 23:55:35 after that the crawling paused for the weekend as the markets are closed. So this change occurred during the weekend.
Upvote: | 100 |
Title: What do you think it's essential math a programmer should know?<p>I've learned higher algebra, calculus, some geometry which was eased by the already known algebra, number theory, probabilities, combinatorics, graph theory, statistics, formal language and automata theory, and that's mostly it. Most of my CS courses in university required some kind of math which we previously did.<p>I can't say I remember all I've learned but I can recall things if needed.<p>It was a thing that programming languages were explained in terms of formal language theory, plus some algebraic constructs. Now a days whenever I read about a programing language, I read about mostly category theory or functional theory and it mostly doesn't make sense.<p>So how hard or how many math domains should one learn to be able to understand all major CS theory?
Upvote: | 104 |
Title: SlickGPT is a light-weight "use-your-own-API-key" ChatGPT client written in Svelte. It offers GPT-4 integration, a userless share feature and other superpowers.
Upvote: | 72 |
Title: On my birthday, the 23rd, while on vacation with my wife, we were about to head out for a special lunch when I received a message asking me to join an all-hands meeting that had been rescheduled. I thought it might be good news from our CEO as they were headed to a convention the next week and thought maybe thy wanted to get it out before they were busy with that. However, two hours later, I found myself jobless as my YC company laid off about 2/3 of its employees via an unexpected Zoom call. I was in the car with my wife, probably 15 minutes passed before I felt like I could drive us home.<p>I'm still in shock, feeling like everything is normal one moment and then suddenly remembering that I'm unemployed.<p>This is the first time I've been laid off in my 26-year work history. Living in rural Indiana, job opportunities are scarce with even the closest gas station being 15 minutes away by car. I'm 20 years into a 4-year degree with only 62 credits to show for it, which doesn't make me an outstanding candidate when there are gobs and gobs with degrees (and usually more varied experience).<p>I'm seeking advice on how to cope with this situation. How can I stay positive and avoid falling into a funk? How can I justify the high cost of COBRA health insurance with limited funds, so my high school teacher wife isn't burdened by my potential medical debt while being our sole provider? How do I fall asleep at night?<p>If you're experiencing this too, I'd love to know how you're handling it. Anyone know of a slack or discord acting as a support group for those recently laid off? Seeing others succeed would bring me welcomed hope I think.<p>I estimate I have about eight weeks before I'll need to take a job at the Taco Bell or McDonald's 16 miles away to make ends meet, ironically Taco Bell erroneously popped up in my LinkedIn job search results for about a month straight earlier this year, it's like they knew I'd need to be a "crew member" in the near future . I'm willing to do it, but I'm afraid I'll be stuck there for years.<p>I've already lost three pounds since the layoff, as stress has killed my appetite and left me feeling sick when I do eat. Anyone else?<p>Any advice or shared experiences are appreciated. I believe camaraderie can provide comfort during tough times like these.
Upvote: | 102 |
Title: Seems like GitHub's struggling at the moment: https://downdetector.com/status/github/<p>I'm getting unicorns.
Upvote: | 323 |
Title: Hey HN, we are Mahmoud and Hammad, co-founders of Play.ht, a text-to-speech synthesis platform. We're building Large Language Speech Models across all languages with a focus on voice expressiveness and control.<p>Today, we are excited to share beta access to our latest model, Parrot, that is capable of cloning any voice with a few seconds of audio and generating expressive speech from text.<p>You can try it out here: <a href="https://playground.play.ht">https://playground.play.ht</a>. And there are demo videos at <a href="https://www.youtube.com/watch?v=aL_hmxTLHiM">https://www.youtube.com/watch?v=aL_hmxTLHiM</a> and <a href="https://www.youtube.com/watch?v=fdEEoODd6Kk">https://www.youtube.com/watch?v=fdEEoODd6Kk</a>.<p>The model also captures accents well and is able to speak in all English accents. Even more interesting, it can make non-English speakers speak English while preserving their original accent. Just upload a non-English speaker clip and try it yourself.<p>Existing text to speech models either lack expressiveness, control or directability of the voice. For example, making a voice speak in a specific way, or emphasizing on a certain word or parts of the speech. Our goal is to solve these across all languages. Since the voices are built on LLMs they are able to express emotions based on the context of the text.<p>Our previous speech model, Peregrine, which we released last September, is able to laugh, scream and express other emotions: <a href="https://play.ht/blog/introducing-truly-realistic-text-to-speech-with-emotion-and-laughter/">https://play.ht/blog/introducing-truly-realistic-text-to-spe...</a>. We posted it to HN here: <a href="https://news.ycombinator.com/item?id=32945504" rel="nofollow">https://news.ycombinator.com/item?id=32945504</a>.<p>With Parrot, we've taken a slightly different approach and trained it on a much larger data set. Both Parrot and Peregrine only speak English at the moment but we are working on other languages and are seeing impressive early results that we plan to share soon.<p>Content creators of all kinds (gaming, media production, elearning) spend a lot of time and effort recording and editing high-quality audio. We solve that and make it as simple as writing and editing text. Our users range from individual creators looking to voice their videos, podcasts, etc to teams at various companies creating dynamic audio content.<p>We initially built this product for ourselves to listen to books and articles online and then found the quality of TTS is very low, so we started working on this product until, eventually we trained our own models and built a business around it. There are many robotic TTS services out there, but ours allows people to generate truly human-level expressive speech and allows anyone to clone voices instantly with strong resemblance. We initially used existing TTS models and APIs but when we started talking to our customers in gaming, media production, and others, people didn't like the monotone robotic TTS style. So we doubled down in training a new model based on the new emerging architectures using transformers and self supervised learning.<p>On our platform, we offer two types of voice cloning: high-fidelity and zero-shot. High-fidelity voice cloning requires around 20 minutes of audio data and creates an expressive voice that is more robust and captures the accent of the target voice with all its nuances. Zero-shot clones the voice with only a few seconds of audio and captures most of the accent and tone, but isn’t as nuanced because it has less data to work with. We also offer a diverse library of over a hundred voices for various use cases.<p>We offer two ways to use these models on the platform: (1) our text to voice editor, that allows users to create and manage their audio files in projects, etc.; and (2) our API - <a href="https://docs.play.ht/reference/api-getting-started">https://docs.play.ht/reference/api-getting-started</a>. The API supports streaming and polling and we are working on reducing the latency to make it real time. We have a free plan and transparent pricing available for anyone to upgrade.<p>We are thrilled to be sharing our new model, and look forward to feedback!
Upvote: | 459 |
Title: I see plenty of hot takes on Twitter about GPT-4 and how its going to kill jobs and give us super powers. What I am not seeing are practical and specific applications outside of "Write code" or "Summarize articles". What examples do you have where someone used GPT-4/AI tech that had a high degree of utility for a specific job to be done? The guy who found his pets illness after the vet gave up is a good specific example applied to a very narrow use case.
Upvote: | 88 |
Title: Hi HN!<p>I was feeling incredibly frustrated with my struggles in context-switching and writing documentation, so I decided to take action and find a solution.<p>I wrote a simple tool for Alfred and Raycast that helps me be more mindful when switching between tasks, which can even prevent some of those switches. Plus, I can jot down quick notes on each task as I go, making it easier to document everything once I’m finished.<p>This tool is really simple with only 5 commands - tsr, tsn, tsl, tsv, tse<p>tsr writes a new record<p>tsn writes a new note<p>tsl shows you the current task<p>tsv builds a static html page displaying a single timeline of the tasks and notes<p>tse opens the directory where records are stored for easy manual editing<p>TSR stores your data as simple csv files in ~/tsr, making it super easy to integrate with other tools or perform your own custom analysis and visualisation magic. The built-in tsv timeline visualisation is rather simple and doesn’t do any analysis (for now at least).<p>It depends solely on Python3 and works offline.
I encourage you to check the python scripts to see how simple they are and potentially adapt them to your own needs.<p>Let me know what you think!
Upvote: | 122 |
Title: This question was asked a few years ago (https://news.ycombinator.com/item?id=22858035) by committed, and I'm curious what it looks nowadays.<p>> How many people on hacker news are running successful online businesses on their own? What is your business and how did you get started?<p>> Defining successful as a profitable business which provides the majority of the owners income.<p>> Have any recent trends affected your business?
Upvote: | 246 |
Title: Hello fellow hackers,
we made a site that gets GPT to answer your question using the info on a webpage you specify or document you upload (e.g., a large textbook .pdf file).<p>Background:
When ChatGPT came out, I had the idea of having it pull answers from my stereo receiver's annoyingly dense 32 page manual. My weekend project prototype proceeded to surprise with great answers—just like what we've all experienced by now. My co-founder thought we should productize it, and make it easy to use online. So here we are with a very early beta! (Try it on a HN thread...)
Upvote: | 86 |
Title: Regex.ai is an AI-powered tool that generates regular expressions. It can accurately generate regular expressions that match specific patterns in text with precision. Whether you're a novice or an expert, Regex.ai's intuitive interface makes it easy to input sample text and generate complex regular expressions quickly and efficiently. Overall, Regex.ai is a game-changer that will save you time and streamline your workflow.
Upvote: | 294 |
Title: I doubt I'm alone in feeling that several of the recent big hype cycles in tech, crypto, nfts, metaverse, now large language models and chat style AIs have all had a certain "snake oil" salesmenesque quality to them. That's not to discredit the underlying achievements and benefits these technologies <i>could</i> bring, but it really does feel like we've experienced a few recent waves of technological mania in which contingents of people have overpromised, overfit, and underdelivered. The discourse around these technologies is quite hyperbolic and free of the more rational, reserved expectations and reasoned discussion I might expect to find around technological developments. Companies seem to be rushing to shove chat AIs into absolutely everything, even where it doesn't make much sense, just to get a blog post out at the detriment of their products. It feels like we're in a sort of gold rush wherein most of the gold is of the fool's variety.<p>Is this a recent thing, or am I simply not remembering similar vaporware hypestorms of the past?
Upvote: | 75 |
Title: Just like the title said - I have always been active & semi-fit but in recent years, I suddenly got struck with terrible muscle soreness. I would often get cramps when waking up, play sport or bad soreness after a light jog, lifting weight or doing any kind of exercise. The soreness won't go away for days, sometimes a week or two which made it really hard to do any kind of exercise and I feel sluggish all the time and this went on for ages without me knowing why and doctor just discard it as "you just need to get used to exercise and it will get better". A year ago, I was just looking for why I get these leg cramps every 2 days, someone suggested taking Magnesium and Vitamin B - This was the turning point.<p>Now, 12 months later and i'm enjoying my life again! I am now able to run up the hills with my kids, wake up without leg cramps at all and the soreness from lifting weight heals within half of a day or a day. I just want to share this if someone is struggling with similar problem as I recently shared it with my dad and a couple of relatives and almost everyone were having great results so it seems like a not very well-known thing. :)
Upvote: | 61 |
Title: As a C++/Networking enthusiast religiously following any new features and quirks in the language and also following several rfcs I have lost complete motivation for self improvement in terms of keeping my skills sharp due to the rapid AI takeover. I had an opportunity to see gpt4 in action on an inhouse product and I'm taken aback, it architects, generates tickets and starts writing code using feature / bug / spike branches for an embedded device a company is working on.<p>It can do almost everything I can do a bit better and I have years and years of domain knowledge, keep ontop of rfc changes, new languages, c++ standards etc, side projects and even occasional leetcode.<p>Oh well, this gold run did run on long enough - Im glad I made a bit of money from the industry but I think all these students going into CS are in for a rude awakening and we're in for a huge shift in this industry.
Upvote: | 72 |
Title: As a software engineer, I've experienced firsthand the challenges of working with the Airtable API.<p>As more non-technical users began using the platform in our growing business, the need for engineering to automate processes and sync data into Airtable grew. However, keeping track of process failures and ensuring that no unresolved failures slipped by was difficult and required significant effort.<p>That's why I created Airwalker, a toolkit that improves the reliability of processes using the Airtable API and helps you correct issues quickly and with minimal effort.<p>Here are some of the features Airwalker offers:<p>* Base schema timeline<p>* Request/response logging<p>* Edit & replay<p>* Custom automatically maintained TypeScript types<p>Airwalker is free to use right now, and I welcome any feedback or suggestions.
Upvote: | 73 |
Title: Hey HN! We’re Taylor, James and Sergio – the founders of Metal (<a href="https://www.getmetal.io/">https://www.getmetal.io/</a>). You can think of Metal as embeddings as a service. We help developers use embeddings without needing to build out infrastructure, storage, or tooling. Here’s a 2-minute overview: <a href="https://www.loom.com/share/39fb6df7fd73469eaf20b37248ceed0f" rel="nofollow">https://www.loom.com/share/39fb6df7fd73469eaf20b37248ceed0f</a><p>If you’re unfamiliar with embeddings, they are representations of real world data expressed as a vector, where the position of the vector can be compared to other vectors – thereby deriving <i>meaning</i> from the data. They can be used to create things like semantic search, recommender systems, clustering analysis, classification, and more.<p>Working at companies like Datadog, Meta, and Spotify, we found it frustrating to build ML apps. Lack of tooling, infrastructure, and proper abstraction made working with ML tedious and slow. To get features out the door we’ve had to build data ingestion pipelines from scratch, manually maintain live customer datasets, build observability to measure drift, manage no-downtime deployments, and the list goes on. It took months to get simple features in front of users and the developer experience was terrible.<p>OpenAI, Hugging Face and others have brought models to the masses, but the developer experience still needs to be improved. To actually use embeddings, hitting APIs like OpenAI is just one piece of the puzzle. You also need to figure out storage, create indexes, maintain data quality through fine-tuning, manage versions, code operations on top of your data, and create APIs to consume it. All of this friction makes it a pain to ship live applications.<p>Metal solves these problems by providing an end-to-end platform for embeddings. Here’s how it works:<p><i>Data In:</i> You send data to our system via our SDK or API. Data can be text, images, PDFs, or raw embeddings. When data hits our pipeline we preprocess by extracting the text from documents and chunking when necessary. We then generate embeddings using the selected model. If the index has fine-tuning transformation, we transform the embedding into the new vector space so it matches the target data. We then store the embeddings in cold storage for any needed async jobs.<p>From there we index the embeddings for querying. We use HSNW right now, but are planning to support FLAT indexes as well. We currently index in Redis, but plan to make this configurable and provide more options for datastores.<p><i>Data Out:</i> We provide querying endpoints to hit the indexes, finding the ANN. For fine-tuned indexes, we generate embeddings from the base model used and then transform the embedding into the new vector space during the pre-query phase.<p>Additionally, we provide methods to run clustering jobs on the stored embeddings and visualizations in the UI. We are experimenting with zero-shot classification, by embedding the classes and matching to each embedding in the closest class, allowing us to provide a “classify” method in our SDK. We would love feedback on what other async job types would be useful!<p>Examples of what users have built so far include embedding product catalogs for improved similarity search, personalized in-app messaging with user behavior clusters, and similarity search on images for content creators.<p>Metal has a free tier that anyone can use, a developer tier for $20/month, and an enterprise tier with custom pricing. We’re currently building an open source product that will be released soon.<p>Most importantly, we’re sharing Metal with the HN community because we want to build the best developer experience possible, and the only metric we care about is live apps on prod. We’d love to hear your feedback, experiences with embeddings, and your ideas for how we can improve the product. Looking forward to your comments, thank you!
Upvote: | 196 |
Title: Requests now appear to redirect to regular reddit.
Upvote: | 98 |
Title: We believe that AI should be fully open source and part of the collective knowledge.<p>The original LLaMA code is GPL licensed which means any project using it must also be released under GPL.<p>This "taints" any other code and prevents meaningful academic and commercial use.<p>Lit-LLaMA solves that for good.
Upvote: | 158 |
Title: Today, billionaires and middle-class everyday people use the same phones. Will that be the case with AIs?
Upvote: | 48 |
Title: Hi Hacker News!<p>My name is Bea, I built a site called Libraria that uses GPT to do a few things<p>1. Let you spin up multiple assistants based on your own documents. You can make it public, private, or protected. It has its own subdomain and landing page.
2. Respond in full markdown always, so it can output images, links, code, and more
3. Let you upload articles on the fly within the Chat, so you can ask it questions
4. Make it embeddable in your site with one line of code
5. Let you update it for fun / with your branding
5. Enable syncing for any URLs you let us scrape, so that you can make sure it's always up to date
6. Let you upload multiple file types<p>I've been working on this for about a month now by myself and you can keep track of my feature updates here: <a href="https://libraria.dev/feature-updates" rel="nofollow">https://libraria.dev/feature-updates</a><p>I would LOVE your feedback on anything, and If you're willing to try it out I'm looking for a few beta users that can provide me more continuous feedback that I would gladly waive the fee for!
Upvote: | 271 |
Title: Did the layoff wave push you to work more hours than expected? Does AI save you time already?
Upvote: | 51 |
Title: Are you seeing pull requests in open source projects that are clearly AI-generated?<p>We recently had someone open a pull request to our open source project and the code and the explanation of the code was clearly AI generated. It was obvious that the code doesn't work and the person had not tested the code. We do not know what the end goal of the person was but we confronted the person and closed the pull requests.<p>Has any other open source projects experienced this? What did you do?
Upvote: | 65 |
Title: Hey HN! Mirrorful (<a href="https://www.mirrorful.com/">https://www.mirrorful.com/</a>) is an open-source developer framework that helps front-end engineers manage their design systems. We’ve been building Mirrorful with the open-source community (<a href="https://github.com/Mirrorful/mirrorful">https://github.com/Mirrorful/mirrorful</a>) and wanted to share our beta with you. Check out our online demo to get the idea: <a href="https://app.mirrorful.com/">https://app.mirrorful.com/</a>.<p>Design systems can be thought of as the “building blocks of your app” which makes me think of Lego bricks. Mirrorful helps you manage your codebase’s Lego bricks and ensure that they are consistent across all of your apps and platforms.<p>We saw as product engineers how hard it is to get code to match Figma mock ups. High-quality design is a competitive advantage, so getting your UI pixel perfect can matter a lot, but is time-consuming and tedious.<p>When we worked for large public companies, we saw that good component libraries help, but engineers are often still dealing with tweaking small design decisions. There are a lot of inefficiencies. We also worked at a small startup and saw what it was like to not have a design system. No design system led to copy pasta code, and days of back-and-forth on simple things like “what hex should i be using for the hover state?”<p>Design systems are tricky to get right. Picking an out-of-the-box solution is easy to begin with, but one day you’ll be cursing yourself due to lack of flexibility (we did!). On the other hand, creating a design system from scratch is super time-consuming even for the best frontend engineers. Mirrorful is our way out of this dilemma.<p>Mirrorful is completely open-source and written in Typescript. We’re starting with basic design elements—commonly called “design tokens” — such as colors, typography, and shadows, but have plans to expand our scope into more complex components.<p>As frontend engineers ourselves, we wanted a tool that lives in code but is visual. It had to be super easy to set up, but also prepare you for scale so you and/or your team don’t end up copy-pasting everywhere. We decided to make it an NPM package (<a href="https://www.npmjs.com/package/mirrorful" rel="nofollow">https://www.npmjs.com/package/mirrorful</a>) that runs a localhost editor and exports out your design tokens into any configuration you want: .js, .ts, .css, .scss, .json. It’s lightweight with no design system lock-in.<p>Our product is completely self-serve: just install our NPM package. If you run Mirrorful locally, a visual dashboard will pop up at localhost:5050 that lets you manage your theme and export various configuration files directly into code.<p>Pricing is similar to other open-source companies—we charge for cloud-hosted features and for premium components.<p>We’ve built open-source/open-core projects before and love interacting with contributors from all over the world. If anyone has any opinions on what we’re building, we’re all ears. Check us out at mirrorful.com and at github.com/Mirrorful/mirrorful and give it a shot!
Upvote: | 134 |
Title: Hey everyone! Kian and Ajay here from Vocode–an open source library for building LLM applications you can talk to. Vocode makes it easy to take any text-based LLM and make it voice-based. Our repo is at <a href="https://github.com/vocodedev/vocode-python">https://github.com/vocodedev/vocode-python</a> and our docs are at <a href="https://docs.vocode.dev">https://docs.vocode.dev</a>.<p>Building realtime voice apps with LLMs is powerful but hard. You have to orchestrate the speech recognition, LLM, and speech synthesis in real-time (all async)–while handling the complexity of conversation (like understanding when someone is finished speaking or handling interruptions).<p>Our library is easy to get up and running–you can set up a conversation in <15 lines of code. Check out our Gen Z GPT hotline demo: <a href="https://replit.com/@vocode/Gen-Z-Phone">https://replit.com/@vocode/Gen-Z-Phone</a> (try it out at +1-650-729-9536).<p>It all started with our PrankGPT project that we built for fun (quick demo at <a href="https://www.loom.com/share/0d0d68f1a62f409eb5ae24521293d2dc" rel="nofollow">https://www.loom.com/share/0d0d68f1a62f409eb5ae24521293d2dc</a>). We realized how powerful voice + LLMs are but that it was hard to build.<p>Once we got everything working, it was really cool and useful. Talking to LLMs is better than all the voice AI experiences we’ve had before. And, we imagined a host of cool applications that people can build on top of that.<p>So, we decided to build a developer tool to make it easy. Our library is open source and gives you everything you need in a single place.<p>We give you a bunch of integrations out-of-the-box to speech recognition/synthesis providers and let you swap them out easily. We have platform support across web and telephony (via Twilio), with mobile coming soon. We also provide abstractions for streaming conversation (this is good for realtime apps like phone calls) and for command-based/turn-based applications (like voice-based chess). And, we provide customizability around how the conversation is done—things like how to know when someone is finished speaking, changing emotion, sending filler audio if there are delays, etc.<p>In terms of “how do you make money” – we have a hosted version that we’re going to charge for (though right now you can get it for free! <a href="https://app.vocode.dev">https://app.vocode.dev</a>) and we're also going to build enterprise products in the future.<p>We’d love for you to try it out and give us some feedback! And, if you have any demos you'd like to see – let us know and we’ll take a crack at building them. We’re curious about your experiences using or building voice AI, what features or use cases you’d love to see, and any other ideas you have to share!
Upvote: | 379 |
Title: Hey HN! I just released go-nbd, a lightweight Go library for effortlessly creating NBD servers and clients. Its a neat tool for creating custom Linux block devices with arbitrary backends, such as a file, byte slice or what I'm planning to use it for, a tape drive. While there are a few partially abandoned projects like this out there already, this library tries to be as maintainable as possible by only implementing the most recent handshake revision and baseline functionality for both the client and the server, while still having enough support to be useful.<p>I'd love to get your feedback :)
Upvote: | 113 |
Title: Hey HN! We're excited to share our new open-source project, Marvin. Marvin is a high-level library for building AI-powered software. We developed it to address the challenges of integrating LLMs into more traditional applications. One of the biggest issues is the fact that LLMs only deal with strings (and conversational strings at that), so using them to process structured data is especially difficult.<p>Marvin introduces a new concept called AI Functions. These look and feel just like regular Python functions: you provide typed inputs, outputs, and docstrings. However, instead of relying on traditional source code, AI functions use LLMs like GPT-4 as a sort of “runtime” to generate outputs on-demand, based on the provided inputs and other details. The results are then parsed and converted back into native data types.<p>This “functional prompt engineering” means you can seamlessly integrate AI functions with your existing codebase. You can chain them together with other functions to form sophisticated, AI-enabled pipelines. They’re particularly useful for tasks that are simple to describe yet challenging to code, such as entity extraction, semantic scraping, complex filtering, template-based data generation, and categorization. For example, you could extract terms from a contract as JSON, scrape websites for quotes that support an idea, or build a list of questions from a customer support request. All of these would yield structured data that you could immediately start to process.<p>We initially created Marvin to tackle broad internal use cases in customer service and knowledge synthesis. AI Functions are just a piece of that, but have proven to be even more effective than we anticipated, and have quickly become one of our favorite features! We’re eager for you to try them out for yourself.<p>We’d love to hear your thoughts, feedback, and any creative ways you could use Marvin in your own projects. Let’s discuss in the comments!
Upvote: | 127 |
Title: Hey Friends,<p>Convoy is an open-source webhooks gateway. Webhooks continue to be hard at scale, and large teams require consistent tooling for inbound & outbound webhooks. Convoy enables developers to securely send, receive and manage millions of webhooks reliably with features like retries, rate limiting, circuit breaking, customer-facing webhook logs, zero downtime secrets rotation and more.<p>Since our initial launch [0], we've learned a lot about our users and made several important improvements we are excited to share:<p>- We are now a webhooks gateway! Akin to API Gateways that sit at the edge of your network to receive all API traffic and route them to the respective microservice, webhooks gateways sit at the edge of your network to receive webhooks from any backend service and route to client endpoints as well as ingest events from multiple providers and route them to the required backend services.<p>- We now have first-class integration with Pub/Sub Systems. Our users wanted increased deliverability guarantees. Your backend services write events to a queue/topic etc. Convoy consumes the queues, creates webhook events and dispatches those reliably to client endpoints. We currently support Amazon SQS and Google PubSub. On our roadmap, we have the following planned - Kafka, RabbitMQ & Nats (In that order)<p>- We switched our backend store to PostgreSQL. This improves the self-hosted experience tremendously. MongoDB was great for storing schemaless data, but was severely lacking in some important features for our users, e.g. transactions don't work on a single node; you need to bootstrap a replica set; also, we wanted to ship updates frequently, but the lack of migrations for schema & data changes slowed us down.<p>- We decided to go the open-core route of OSS monetisation because it offered us a good balance to serve our community and make enough money to run the company. Like GitLab, we hope to be good stewards of our community edition. Since our enterprise edition is simply a wrapper around the community edition with enterprise features like RBAC, Audit Logs etc., we are properly incentivised to continue making it excellent.<p>- Our Cloud platform is in the private alpha stage. Please contact us at [email protected] to gain access!<p>Our mission is to serve hobbyist developers all the way to the most ambitious teams on the planet with a consistent and easy-to-use webhooks gateway for asynchronous communication on the internet.<p>We welcome you to try it out using our getting started at <a href="https://github.com/frain-dev/convoy#installation-getting-started">https://github.com/frain-dev/convoy#installation-getting-sta...</a>. Share with us your webhook horror stories and give us feedback.<p>[0]: <a href="https://news.ycombinator.com/item?id=30469078" rel="nofollow">https://news.ycombinator.com/item?id=30469078</a>
Upvote: | 119 |
Title: It's been closed source for a long time while I worked on it on and off as a hobby research project, but yesterday the repo was made public for the first time under the MPL 2.0 license.<p>A feature reel showing its capabilities can be found here:
<a href="https://ktstephano.github.io/rendering/stratusgfx/feature_reel" rel="nofollow">https://ktstephano.github.io/rendering/stratusgfx/feature_re...</a><p>A technical breakdown of a single frame can be found here:
<a href="https://ktstephano.github.io/rendering/stratusgfx/frame_analysis" rel="nofollow">https://ktstephano.github.io/rendering/stratusgfx/frame_anal...</a><p>It's still in a very beta state (bugs and instability expected), but I felt like it was a good time to make it public since a lot of its core features are mostly presentable. I plan to continue working on it in my spare time to try and improve the usability of the code.<p>Two main use cases I could see for it:<p>1) People using it for educational purposes.<p>2) People integrating it into other more general purpose engines that they're working on since Stratus is primarily a rendering engine. Any extensions to the rendering code that are made public would then further help others.<p>So I think it will remain very niche but I'm hoping it will still be helpful for people in the future.
Upvote: | 483 |
Title: Hi Hacker news !<p>I’m Julien and I built an alternative CLI for Git : gut.<p>Even if I haven’t been coding for a long time (I’m in the first year studying computer science), I’ve always found git to be frustrating.
The command naming is inconsistent and git lets you easily shoot yourself in the foot.<p>I made gut, another git porcelain, to solve these issues.<p>It provides a consistent naming of command. To do so, syntax is based on subcommands. For example, to delete a branch, run gut branch rm rather than git branch -d, same to delete a remote (gut remote rm) and so on.<p>Gut also prevents you from shooting yourself. It provides nice defaults and always prompt you before doing something destructive.
Also, it won’t allow you to rewrite the history if it has been pushed to the remote. Creating commits in detached head is also prohibited.<p>Finally, git was made when GitHub and others didn’t existed yet. To diff commits, gut opens the compare view in the browser. And to merge a branch, gut opens a pull request.<p>I have been working on this project for the past few months and I am happy to be able to share it.<p>I hope you’ll like it. Any suggestions is welcome !<p>Repo: <a href="https://github.com/julien040/gut">https://github.com/julien040/gut</a>
Upvote: | 314 |
Title: Greetings!<p>YakGPT is a simple, frontend-only, ChatGPT UI you can use to either chat normally, or, more excitingly, use your mic + OpenAI's Whisper API to chat hands-free.<p>Some features:<p>* A few fun characters pre-installed<p>* No tracking or analytics, OpenAI is the only thing it calls out to<p>* Optimized for mobile use via hands-free mode and cross-platform compressed audio recording<p>* Your API key and chat history are stored in browser local storage only<p>* Open-source, you can either use the deployed version at Vercel, or run it locally<p>Planned features:<p>* Integrate Eleven Labs & other TTS services to enable full hands-free conversation<p>* Implement LangChain and/or plugins<p>* Integrate more ASR services that allow for streaming<p>Source code: <a href="https://github.com/yakGPT/yakGPT">https://github.com/yakGPT/yakGPT</a><p>I’d love for you to try it out and hear your feedback!
Upvote: | 287 |
Title: Hi HN, we’re building Noya (<a href="https://noya.io">https://noya.io</a>), a product design tool that lets everyone design like a designer. Here’s a video: <a href="https://player.vimeo.com/video/797476910" rel="nofollow">https://player.vimeo.com/video/797476910</a>.<p>Getting from a product idea to a design is too hard. Wireframing tools (e.g. Balsamiq) are limited and their output is too low-fidelity to get people excited about the idea. On the other hand, high-fidelity design tools (e.g. Figma) require advanced skills and are better for tweaking fine details than hashing out a big picture. Meanwhile, companies have an insatiable need to make new screens and change existing ones, and hiring more designers is time-consuming and expensive.<p>David and I were on the design tools team at Airbnb when we realized there’s another solution: let designers encode their knowledge (e.g. design rules and components) into a tool, then let non-designers (PMs, marketing, engineers) use that tool to make new screens and features. This helps remove design as a bottleneck for a lot of product development. We built such tools at Airbnb, and with Noya we’re building them for product teams everywhere.<p>Current design tools are too freeform for non-designers to design great products. They let you do anything, like draw rectangles and move text blocks anywhere, so it's easy to mess things up, introduce inconsistencies, and so on. With Noya, designers set up "guardrails" in the form of a design system (rules and components for a company's design), then non-designers work within those constraints. This makes it harder to mess up and quicker to build something that fits in with your product. Footguns begone!<p>It all starts with wireframing, i.e. drawing a minimal layout that shows the elements that would exist on the screen. Noya combines wireframes and design systems to generate high-fidelity designs and code. If you have an idea for a user interface, you can use Noya to quickly wireframe that idea by clicking and dragging to place blocks for each element of your user interface. For each block, choose a type and provide any content that goes inside it.<p>Based on the rules and component library of the design system you select, Noya automatically turns your wireframe into a high-fidelity design. This design can be exported to design files, to Figma or Sketch, or to React code.<p>(If you’re curious what the React code looks like, try exporting and take a look! There’s plenty of room for improvement, especially around responsive layouts, but we think it’s a reasonable starting point. The code export is configurable on a per-design-system basis so that it’s closer to a company’s preferred standards).<p>Most tools in this space are optimized for either low-fidelity wireframes that are quick to create (e.g. Balsamiq), or high-fidelity output that’s slow to create (e.g. Figma). We think there’s a gap in the market for a great wireframing tool that produces a high-fidelity output quickly. For example, two-thirds of Figma users are non-designers. While there are many valid reasons for a non-designer to use Figma, there’s often a lot of upfront effort required to learn the tool and set up components. We think there should be a lower-effort way for non-designers to create high-fidelity designs.<p>Based on feedback from our Show HN (<a href="https://news.ycombinator.com/item?id=34848583" rel="nofollow">https://news.ycombinator.com/item?id=34848583</a>), we’ve improved our onboarding, revamped our entire block library, and added a documentation reference for each block type. The docs are interactive, so you can play with our editor without signing up: <a href="https://noya.io/app/docs">https://noya.io/app/docs</a>.<p>Startup founders and PMs have used Noya to build landing pages, dashboards, and other flows of their apps. We have some templates on our site (<a href="https://noya.io/templates">https://noya.io/templates</a>) that give examples of what people commonly build in Noya. We use Noya ourselves, and have been surprised by how frustrating it was to go back to existing design tools after using even the earliest prototypes of Noya. Existing design tools just aren’t built for the comprehensive design systems that products are based on today.<p>The source code is available here (though largely undocumented & unlicensed): <a href="https://github.com/noya-app/noya">https://github.com/noya-app/noya</a>. We’re not focused on growing our open source community or supporting other design tool builders just yet, so we haven’t published our packages (renderer, canvas, etc) to npm, but we’re planning to use the Apache V2 license. We make money by offering a paid subscription.<p>Noya currently supports one design system based on Chakra UI. We’re adding additional design systems soon, including Material Design, as well as the ability to import custom design systems. If your company has an open source design system and you’re interested in trying Noya with it, we may be able to integrate your design system for free. Get in touch if you’re interested - [email protected].<p>We’d love for you to try Noya and let us know what you think! It’s still very much an MVP and all kinds of feedback are welcome.
Upvote: | 241 |
Title: Hey hacker news, we launched a few weeks ago as a GPT-powered chatbot for developer docs, and quickly realized that the value of what we’re doing isn’t the chatbot itself. Rather, it’s the time we save developers by automating the extraction of data from their SaaS tools (Github, Zendesk, Salesforce, etc) and helping transform it to contextually relevant chunks that fit into GPT’s context window.<p>A lot of companies are building prototypes with GPT right now and they’re all using some combination of Langchain/Llama Index + Weaviate/Pinecone + GPT3.5/GPT4 as their stack for retrieval augmented generation (RAG). This works great for prototypes, but what we learned was that as you scale your RAG app to more users and ingest more sources of content, it becomes a real pain to manage your data pipelines.<p>For example, if you want to ingest your developer docs, process it into chunks of <500 tokens, and add those chunks to a vector store, you can build a prototype with Langchain fairly quickly. However, if you want to deploy it to customers like we did for BentoML ([<a href="https://www.bentoml.com/](https://www.bentoml.com/)" rel="nofollow">https://www.bentoml.com/](https://www.bentoml.com/)</a>) you’ll quickly realize that a naive chunking method that splits by character/token leads to poor results, and that “delete and re-vectorize everything” when the source docs change doesn’t scale as a data synchronization strategy.<p>We took the code we used to build chatbots for our early customers and turned it into an open source framework to rapidly build new data Connectors and Chunkers. This way developers can use community built Connectors and Chunkers to start running vector searches on data from any source in a matter of minutes, or write their own in a matter of hours.<p>Here’s a video demo: [<a href="https://youtu.be/I2V3Cu8L6wk](https://youtu.be/I2V3Cu8L6wk)" rel="nofollow">https://youtu.be/I2V3Cu8L6wk](https://youtu.be/I2V3Cu8L6wk)</a><p>The repo has instructions on how to get started and set up API endpoints to load, chunk, and vectorize data quickly. Right now it only works with websites and Github repos, but we’ll be adding Zendesk, Google Drive, and Confluence integrations soon too.
Upvote: | 63 |
Title: Hey HN! Andrew and Ani here, co founders of Baseplate (<a href="https://www.baseplate.ai">https://www.baseplate.ai</a>). We're building a unified backend for LLM apps, where you can manage your data, prompts, embeddings, and deployments. Demo video here: <a href="https://www.loom.com/share/97bc388f6b6c4c95bf6c51c832bfd8a5" rel="nofollow">https://www.loom.com/share/97bc388f6b6c4c95bf6c51c832bfd8a5</a>.<p>Most LLM apps can be built by properly integrating LLMs with a knowledge base consisting of domain-specific or company-specific data. The scope of this knowledge base can change based on the task- it can be something as narrow and static as your API docs or as broad and fluid as meeting transcripts from your customer support calls.<p>To effectively use their data, most teams need to build a similar stack—datasource integrations, async embedding jobs, vector databases, bucket storage for non textual data, a way to version prompts, and potentially an additional database for the text data. Baseplate provides much of the backend for you through simple APIs, so you can focus on building your core product and less on building common infra.<p>At my previous role at Google X, I worked on building data infrastructure for geospatial data pipelines and knowledge graphs. One of my projects was to integrate knowledge graph triples with LaMDA, and I discovered the need for LLM tooling after using one of Google's initial prompt chaining tools. Ani was a PM at Logitech, shipping products in their Computer Vision team, and at the same time building side projects with GPT-3.<p>The core of Baseplate is our simplified multimodal database, which allows you to store text, embeddings, data, and metadata in one place. Through a spreadsheet-style interface, you can edit your vectors and metadata (which is surprisingly complex with existing tools), and add images that can be returned at query time. Users can choose between standard semantic search or hybrid search (weighted keywords/semantics for larger, more technical datasets). Hybrid search on Baseplate utilizes two open-source models that can be tuned for your use case (instructor & SPLADE). Datasets are organized into documents, which you can keep in sync through our API or through the UI (this way you can keep your datasets fresh when ingesting data from Google Drive/Notion/ etc).<p>After your datasets are set up, we have an App Builder where you can iterate on prompts with input variables, and create context variables that pull directly from a dataset at query time. We give you all the knobs and dials, so that you can configure exactly how your search is performed and how it is integrated with your prompt.<p>When you're satisfied with an app configuration, you deploy it to an endpoint. All you need is a single API call and we'll pull from one (or multiple) datasets in your app and inject the text into the prompt. We also return all the search results in the API response, so you can build a custom UX around images or links in your dataset. Endpoints have built in utilities for human feedback and logging. With GPT-4 being able to take images as input, we will soon be working on a way to pipe images from your dataset directly to the model. And all of these tools are in a team workspace, where you can quickly iterate and build together.<p>We just started offering self-serve sign up, and our pricing is currently $35/month per user for our Pro plan and $500/team on our Team plan. Feel free to sign up and poke around. We'd love to hear feedback from the community, and look forward to your comments!
Upvote: | 177 |
Title: Neural networks! Bah! If I wanted a black box design that I don't understand, I would make one! I want rules and symbolic processing that offers repeatable results and expected outcomes!<p>...and maybe there's still a place for that.<p>But for someone who has been possibly foolish and ignoring neural network-oriented AI, where's the best place to start learning?<p>Go back to the basics of recurrent neural networks, deep learning texts, and so forth? Or is there a shortcut into the hip and popular transformer-based technology at this point?
Upvote: | 58 |
Title: Hey team, I'm Arjun! Builder at flowy.live.<p>So - it's quite simple. It's a walkie-talkie with a way to play back the last 24 hours. A new medium of communication for active teams.<p>We have released for Mac and Windows.<p>Any feedback or thoughts would mean the world to us!<p>Open to a feedback session? <a href="https://calendly.com/arjun-flowy/onboarding" rel="nofollow">https://calendly.com/arjun-flowy/onboarding</a><p>Appreciate all of you,
Arjun Patel
[email protected]
Upvote: | 43 |
Title: An odd question, but something I've never really considered until lately.<p>How does one become smarter? I'm in my early 30s right now. I can feel my brain waning here and there--things that I used to pick up quickly are now taking a long time to settle in. I'm wondering why that is.<p>That seems to concern maintaining one's intelligence. But what if one wanted to improve it? How can I go about doing that?<p>Thank you.
Upvote: | 81 |
Title: Hi HN!<p>Sharing Random Airport<p>Inspired by RandomStreetView (which I find weirdly addictive), and a passion for air travel.<p>Probably not for everyone, but I hope some of you find it interesting!<p>Needless to say, open to feedback!<p>Enjoy clicking!<p>Further reading:<p>TECH: It's Build in React, NodeJS, with a Notion DB. The code is public on Github. It is spaghetti though ... Especially open to feedback here.<p>DB: The db is publicly available (and editable), I can add the link in comments if anyone would like to have a look .<p>KNOWN ISSUES: I would like to improve the design, pic loading performance and quality of (some) pics.
Upvote: | 175 |
Title: Good to be back on HN with all-new CoScreen, a little more than 3 years after it launched over here!<p>With CoScreen 5.0, you can now share your windows from multiple displays at the same time, a long standing request by our most avid users and impossible in other apps. It also has a lightning-fast, Rust-based window compositing, scaling, and streaming engine now.<p>CoScreen was always meant to be different so that you and your team can share your screens simultaneously and multi-directionally, and to be able to control what is being shared. We saw it as a natural extension and closely coupled with your OS — instant, fast, and seamless. A better way to pair program, debug tough incidents, or jam on great ideas by sharing multi-modal information like code, commands, graphs, or logs.<p>All that made a lot of sense conceptually but to be frank, it was hard to get it right. Now a part of Datadog and with major parts of our app rewritten in Rust, we feel we’re closer than ever.<p>Here’s what pair programmers liked about CoScreen, so we made it even better:
- High definition code sharing: Windows are video-streamed in real-time at their native resolution whenever possible. You never have to search for your IDE anymore or be anxious to share the wrong window.
- Multi-directional collaboration: You can share, while Alice shares, while Bob shares. Side-by-side, across multiple displays. With built-in crisp audio and video chat.
- 60FPS+ super smooth mouse pointers. Type, click, and draw on any shared window as if it was your own.<p>What some of you did NOT like, so we fixed it in CoScreen V5:
- CPU utilization and latency have been reduced drastically as various parts of our desktop client are now implemented in Rust, building on crates such as cxx, rust-skia, iced, as well as Neon for our native remote control plugins.
- No more accidental clicking into remote windows through the new remote window toggles.
- You’re no longer bound by your displays, can share windows from multiple of them at the same time and even move them across displays while sharing without stopping.
- You’ll also soon be able to join meetings from your browser from any platform.<p>CoScreen runs on macOS (x64 and Apple Silicon), Windows, soon also on the web and is currently free. We’re planning to charge for larger teams and enterprise features in the future. Hopefully - finally - we’ll also have a Linux version one day. Tell us if you need it urgently and if you have any other requirements!
Upvote: | 350 |
Title: Hello everyone, I’m Nishad, cofounder at Inri (<a href="https://www.goinri.com/">https://www.goinri.com/</a>), an investment platform for Indian expats to diversify capital in India, without any taxation or repatriation hassles. Super excited to be launching on HN, the platform that I personally nerd on and has helped me discover some of my favorite products early.<p>My cofounder Hemant (@hemantgangolia) has been living in the US since 6 years and has tried multiple times to invest in India. He couldn't go ahead with the process because bank account opening was a challenge and the process after that wasn't clear. There is complexity around taxation, moving money back and a general lack of clarity on where to invest.<p>Even though there are currently over 30 million Indian expats globally, there are no digital, full-service solutions offered for them specifically. US solutions focus only on US investments and Indian solutions focus only on Indian customers—this makes the India <> US belt underserved. Indian banks with overseas branches offer these services, but investors don’t trust these banks because of conflict of interest. There are other small players in the market but they expect customers to do their own service, which is not what customers want, based on our learnings.<p>We found a lot of other Indian expats in the same situation as us - wanting to invest in India for financial and emotional reasons, but unable to do so easily or consistently. We wished a simple product existed to solve for this online. We couldn’t find one, so we built it.<p>Inri is like Wealthfront for your India investments. It is a web platform that offers curated mutual fund portfolios and solves for all tax compliance and repatriation needs online. We have curated the funds based on performance, history and compliance based on your resident country, and funds are selected based on your risk preference, similarly to Wealthfront. We facilitate investments through central platforms like National Stock Exchange and show the holdings via a dashboard on the platform once the investments are done. Additionally, we do this while making tax compliance and repatriation easy.<p>Just log in to the platform, upload your documents, approve the emails with details confirmation and you’re sorted. Currently, only people with a PAN card (Indian financial identity document) can invest.<p>Our early customers are all people who have tried doing this before, but gave up because of all the hassle. The fact that they could get invested in just 2 days vs weeks for traditional alternatives was unheard of to them.<p>We know this is a bit of a niche product for HN but we also know that there are quite a few Indian expats like us here, who could make use of it—and we hope it makes interesting enough reading for everybody else. We would love to get your feedback! Thank you!
Upvote: | 146 |
Title: Cadseer is in the same genre as: solidworks, inventor, freecad etc... Cadseer is alpha software, so lots of missing features, bugs, crashes, file incompatibilities etc...<p>About me: I spent 15 years designing stamping dies on a high end cad system. During that time, I also developed and marketed extensions/plugins toward that cad system. Through that experience, I came disillusioned with 'vendor lock-in' and proprietary software in general. I basically retired, took the vow of poverty and moved all my computing to open source. I have since, and continue to, try and improve the open source cad environment.
Upvote: | 102 |
Title: Hi HN!<p>I made a tool that autogenerates simple, high-level explanations of concepts and organizes them in a somewhat university course-like structure so that it's easier to see how things are structured. Currently it has about 20,000 concepts on a range of topics but that's just what I generated so far, it should work with more obscure topics in the future.<p>I love learning about random topics where I don't have a good background in like history or linguistics, but it's hard to figure out what topics there (you don't know what you don't know) are in certain fields and what they are even about, so this was a way to get the high level idea about random things that I wanted to know about.<p>It also only uses the information in the GPT model at the moment, so obviously information can't be trusted completely and you should definitely double check any information you read here by Googling. I'm thinking of doing the Bing Chat approach for the next version and adding references, but don't have that yet<p>Hopefully someone else finds this useful even if it's not perfect!
Upvote: | 61 |
Title: Ever since the DPReview closure announcement <a href="https://news.ycombinator.com/item?id=35248296" rel="nofollow">https://news.ycombinator.com/item?id=35248296</a> we were thinking how to preserve the 25 years of valuable DPReview camera data. Archive.org has been great, but it's not usable by the general public.<p>The best way to keep it safe going forward, is to have the community own it, so we open sourced it: <a href="https://github.com/open-product-data/digital-cameras">https://github.com/open-product-data/digital-cameras</a><p>I'm aware of a number of attempts to make product data open-sourced, but none have the power of the photo geeks behind it :)<p>Thoughts or ideas? + really looking for some contribution love.
Upvote: | 177 |
Title: Hey Hacker News!<p>My name is Collin, 18 years old and doing a gap year after finishing high school last year.<p>This was my first real project after starting to learn web development around 5 months ago.<p>I came up with this idea as it was a real pain for me to find other people from my country and especially my age, learning and taking online courses about the same stuff online. Lots of these online courses include their own discord communities and forums, but I still found it very hard to connect with other people in there.<p>Thats why I built Coursemate.<p>I would love to get your feedback on it! :)
Upvote: | 283 |
Title: Hello everyone,<p>I am the founder of a blogging platform called Feather.so.<p>People can sign up and create their own blogs using Feather.<p>Now, with OpenAI releasing their API, they made using AI so accessible for someone like me. So I wanted to add a chatbot functionality to my customer blogs. Basically, I wanted to automatically create a chatbot for each of my customer blogs. That chatbot will be trained on the content on their blog.<p>When I set out to do this using Open AI, I thought I could do this for every website, not just for my customer blogs.<p>So I ended up creating an entirely new product called SiteGPT.ai so that it can be used on any website.<p>The workflow works like this. People login the platform, they enter their website url, and click on a button to start training. Then I start creating a chatbot and train the chatbot will all the content on the website that the user enters.<p>That chatbot now knows everything about that website. It can answer any questions related to that website.<p>I have also added a demo chatbot at the bottom right of the sitegpt.ai website. That chatbot is trained on the content of SiteGPT.ai. So it can answer any questions related to its own website.<p>Please try it out and let me know if you have any feedback. I am also happy to take any other technical questions you may have.<p>Thanks.
Upvote: | 118 |
Title: It seems that Google is suffering from gpt influence, and started putting gpt everywhere.<p>I was searching certain words and the interface drastically changes based on what they think of the word. Then again, the auto completion is apparent hack of some gpt models in a rush.<p>It really used to be cool and stable, at least for me. Deep learning is not a magic hammer.
Upvote: | 57 |
Title: My dad, age 85 and a lifelong technophobe bordering on tech-hostile, began using a MacBook Air and iPhone regularly a few years ago. His activities are basic: text me and family, email, photos, read about art and music via sites found on Google, YouTube for how-to videos about specific art techniques, language learning with Duolingo and help from Google Translate, and some very infrequent purchases mostly on eBay and Amazon. He still struggles with the most basic UIs but he gets things done, it’s been great to see.<p>But he keeps getting scammed. He usually recognizes it after the fact, he’ll fill out a phishing form and call me right away, “I did it again.” He always feels embarrassed. They find him through emails and text messages. He ignores many of them (I know because he tells me, “Another one came through!”) but there seem to be so many that some get him.<p>I had to help him with something on his phone the other day and when I went to open a new tab in Mobile Safari, I saw no fewer than six different scam pages up. Fake Amazon, fake UPS, fake credit card. It was frightening. I’m worried he’s inching towards something catastrophic like sharing bank account information. It’s also making him afraid to use technology. He doesn’t want a credit card anymore, he’s so tired of having to change the number.<p>I don’t know what to do. He’s found so much independence thanks to technology, he’d be isolated if he stopped using it. He struggles with the most basic user interfaces, details that I take for granted are invisible to him, so I don’t think he’s likely to learn all the tricks of scammers. I can’t look over his shoulder all the time.<p>Does anyone have any advice for this? Any experience?
Upvote: | 115 |
Title: Hey HN! We’re Ioannis & Zaf, building Algora.io to help open source projects reward their contributors & grow their communities.<p>1 min demo: <a href="https://twitter.com/algoraio/status/1641560954746839042" rel="nofollow">https://twitter.com/algoraio/status/1641560954746839042</a><p>The problem: paid contributions in open source are scarce, low trust & high friction<p>Our solution: we built an app that streamlines open source bounties on Github<p>Our 1st customer was Remotion.dev (15.6k stars, Typescript/React) in November 2022, whose feedback helped us ship our Github app & iterate through our bounty workflow. To date, Remotion.dev has rewarded 17 open source bounties: <a href="https://github.com/remotion-dev/remotion/issues?q=label%3A%22%F0%9F%92%8E+Bounty%22+is%3Aclosed">https://github.com/remotion-dev/remotion/issues?q=label%3A%2...</a><p>Since then, we’ve been fortunate to also onboard Cal.com (17.6k stars, Typescript/Next), IHP (3.9k stars, Haskell), Qdrant.tech (5.4k stars, Rust), erxes.io (2.8k stars, Typescript) and shuttle.rs (YC S20, 2.1k stars, Rust).<p>OSS contributors in the US, Europe (Germany, France, Norway), Canada, Nigeria, India, Egypt, UAE, Brazil, Colombia, Philippines and Australia have already earned bounties with Algora — we hope this list keeps growing!<p>We also started a COSS founder interview series to share lessons & advice for building open source companies: <a href="https://youtube.com/@algora-io">https://youtube.com/@algora-io</a><p>We are really excited to hear your feedback/questions and connect further: our emails are [email protected] & [email protected]. Thank you!
Upvote: | 111 |
Title: Hello all :) I made this BBS for fun and thought you all would enjoy it. It's not perfect, but it's been a fun exercise!<p>see it: <a href="https://asciinema.org/a/Emg6SWrXMV6cehfQxrw1GRu75" rel="nofollow">https://asciinema.org/a/Emg6SWrXMV6cehfQxrw1GRu75</a><p>try it:<p><pre><code> ssh -p 2223 shhhbb.com
</code></pre>
host it: <a href="https://github.com/donuts-are-good/shhhbb/releases/latest">https://github.com/donuts-are-good/shhhbb/releases/latest</a><p>why?<p>Every year I challenge myself in some new way, this year it is to push one project per week. You might recognize my static site generator [0] or my releaser for go [1] from previous posts as one of these weekly projects. If you want to join me in doing this, it's been a blast and I highly recommend it! Maybe we can chat on the bbs about it :)<p>[0] - bearclaw tiny static generator - <a href="https://github.com/donuts-are-good/bearclaw">https://github.com/donuts-are-good/bearclaw</a><p>[1] - release.sh release builder for go - <a href="https://github.com/donuts-are-good/bearclaw">https://github.com/donuts-are-good/bearclaw</a><p>need:<p>I'd love a few co-conspirators, or even some new friends for the bbs software or the bbs itself. A lofty nice-to-have goal is meeting a few other similarly motivated people to conspire with on a weekly basis. If that's you, drop me a line!<p>goals:<p>I have about half of an admin interface endpoint pushed up, which I'd like to finish. I realized mid-commit that I'd lacked some other material in a previous commit, and in pushing it up, I ended up pushing half the admin feature. Stay tuned for that.<p>I'd also like to implement a feature where I surrender the socket to a new process so I can upgrade the binary without dropping all of the connected users. I'm not sure on the feasibility of this, but all the words fit together, so it'll be tried.
Upvote: | 188 |
Title: I'm looking for some "What Would You Do?" advice from the founder and investor community. Generally looking for guidance on what I should be negotiating for and what type of downside protection is fair, in any?<p>Financial Overview<p>- At our peak (Series C) we were valued at $1B post money - $187M raised by the company
- $250M preference stack (liquidation prefs + debt)
- I own ~27% of the business, and so does my co-founder. My co-founder is no longer part of the business day-to-day, he's on the board
- We will end the year at $130M in annual revenues and burning ~$925K/month, we are a low gross profit margin business ~15-20%<p>Personal Overview
- I live in New York City, it's been ~ 10 years that I have been on this journey, bootstrapped with no salary for about 2.5 years and pretty modest salary for most of the years, up until a couple years ago when I started getting "market rate"
- I have not taken any chips off the table via any secondary sale up to this point, never really got the chance
- Today, my base salary is $175K, no bonus, no stock refresh since the founder grant, fully vested stock, no severance, no indemnification. (Mainly because I just haven't asked for new comp b.c company is in a difficult spot)
- In comparison, I hired a critical new C-Suite exec (COO) and gave her $285K/year, 75-150% annual target bonus, $150K sign on bonus, $200K funding closed bonus, 1-year severance, and a 4% equity stake in the business
- I am starting to see some worrying signs of real burnout now (for myself)<p>State of the Union
- We have < 6 months of runway based on our current burn, but majority of the money left is debt now
- I am about to close on $30M of additional financing this month but the terms will decimate common stock holders (my 27% stake is likely going to go down to < 10%)
- Still negotiating, but will likely have to give up board control so I could get fired immediately after financing is closed
- We are in parallel exploring strategic M&A alternatives this year, but in the current market condition our business is likely worth less than the preference stack, in which case common stock is wiped out anyways and I get nothing.
- Lastly, I'm not confident on the current strategy to grow out of the current problem over the next couple years, so there's a lot of risk in our growth for 2024 and beyond. The strategy I would want to run as CEO is likely not something my COO or board will want to get behind and frankly I don't know if I have the energy to do it myself. It would require 24-36 months to transition the business to the new world while watching old revenue decline and waiting for new higher margin revenue build up in parallel to offset the declining revenues in the core business.<p>My #1 priority is to get the financing closed for the business. But from there, I'm really confused and don't know what to do next. Doesn't seem to make sense for me to work my ass off just to return money to investors and have nothing for myself. But I do understand that it is my fiduciary obligation as the Founder/CEO, but maybe for the first time in 10 years I'm asking the question "What's in it for me?"<p>WWYD?
Upvote: | 122 |
Title: Hey,<p>I work at a service oriented dev company and from time to time we have some bench time to utilize. I'd like to keep people engaged on projects and ideally exposed them to best practices as much as possible.<p>What's the best way to find places to "donate" our time to certain charities or non-profit organizations?<p>Right now people are going out to our most used third party open source software repos and actively contributing there, but I'd like to do more of a full on apps.<p>Thanks
Upvote: | 94 |
Title: Please lead with either SEEKING WORK or SEEKING FREELANCER,
your location, and whether remote work is a possibility.
Upvote: | 46 |
Title: Please state the location and include REMOTE, INTERNS and/or VISA
when that sort of candidate is welcome. When remote work is <i>not</i> an option,
include ONSITE.<p>Please only post if you personally are part of the hiring company—no
recruiting firms or job boards. One post per company. If it isn't a household name,
explain what your company does.<p>Commenters: please don't reply to job posts to complain about
something. It's off topic here.<p>Readers: please only email if you are personally interested in the job.<p>Searchers: try <a href="https://hnhired.fly.dev" rel="nofollow">https://hnhired.fly.dev</a>, <a href="https://kennytilton.github.io/whoishiring/" rel="nofollow">https://kennytilton.github.io/whoishiring/</a>,
<a href="https://hnjobs.emilburzo.com" rel="nofollow">https://hnjobs.emilburzo.com</a>, <a href="https://news.ycombinator.com/item?id=10313519" rel="nofollow">https://news.ycombinator.com/item?id=10313519</a>.<p>Don't miss these other fine threads:<p><i>Who wants to be hired?</i> <a href="https://news.ycombinator.com/item?id=35424805" rel="nofollow">https://news.ycombinator.com/item?id=35424805</a><p><i>Freelancer? Seeking freelancer?</i> <a href="https://news.ycombinator.com/item?id=35424806" rel="nofollow">https://news.ycombinator.com/item?id=35424806</a>
Upvote: | 418 |
Title: Share your information if you are looking for work. Please use this format:<p><pre><code> Location:
Remote:
Willing to relocate:
Technologies:
Résumé/CV:
Email:
</code></pre>
Readers: please only email these addresses to discuss work opportunities.
Upvote: | 183 |
Title: Hi HN! I’m Joseph, and along with Arpan and Bailey we are the founders of OutSail Shipping (<a href="https://outsailshipping.com/">https://outsailshipping.com/</a>). We’re building a sail the size of a 747 that rolls up into a shipping container. When deployed, it will generate thrust from the wind to reduce the fuel consumption of a cargo ship. An array of these devices will reduce fuel consumption on ships by up to 20%. These sails are easily stowed and removed to cause no interference with cargo operations. Here’s a short video showing our prototype: <a href="https://www.youtube.com/watch?v=VUpVqzpym54">https://www.youtube.com/watch?v=VUpVqzpym54</a>.<p>Sails powered ships for millennia; but then the convenience of energy-dense fuels displaced sails. As ship speeds eventually exceeded wind speeds, the consensus became that sails had no place in shipping and were relegated to hobbyists and sport. Fast forward a century and a half, and maritime shipping, like all other industries, is facing a reckoning to mitigate the greenhouse gasses produced by their activities.<p>The International Maritime Organization (IMO) has introduced new regulations which use a vessel’s Carbon Intensity Indicator (CII) to grade ships. This grading scale becomes more aggressive over time, and any ship with a poor grade must take corrective action. The corrective actions can be as non-invasive as reducing speed (aka: slow steaming) or as extreme as a retrofit to use a different, cleaner fuel source. This costs millions and takes a ship out of commission for months, and it’s difficult to ensure your (now more expensive) fuel is available at every port of call. Ship owners are hedging their bets that slow steaming will dominate their future, with ship order books full to reflect the increased capacity needed when containers take 20% longer to cross the ocean.<p>Or option three. There is sufficient wind on the ocean to power the entire shipping industry, if you’re willing to grab it. Wind Assisted Ship Propulsion (WASP) devices can be used as a corrective action to improve a vessel’s CII rating, without reducing ship speed or changing the route. In other words, a return of sails.<p>We are hardware engineers with over two decades of experience between us, working at Tesla, SpaceX, JPL, Relativity, and some startups. The idea for OutSail came from Arpan and Joseph getting coffee after work one day. When we asked each other “What would you do if you weren’t building satellites?” maritime cargo came up from both sides; Arpan from having studied the industry for opportunities to reduce emissions, and Joseph from a love of hydrodynamics and maybe too many sea-shanties. Bailey and Arpan, meanwhile, had been looking at working on bicycling infrastructure. What brought the three of us together was actually a Dungeons & Dragons game where we realized we made a good team! We settled on OutSail as a good fit for our hardware hacking mentality, trading in our druids staffs for spanners.<p>Aerodynamically, sails are simply vertical wings. Wind blowing across the vessel causes the sail to generate lift and drag, and the resultant vector has some forward component to pull the ship through the water. However, if the wind comes from an angle too close to the direction of travel, there is no thrust. As an added complication, the sail only sees the <i>relative</i> wind. If the ship travels faster, the wind will appear to come from closer and closer to the direction of travel, even if the true wind is coming from perpendicular to your course! Despite this, standard sails can still produce forward thrust as long as the wind is at least 20 degrees off from directly in front of the vessel. This is how our sails can still save power, even on a fast moving vessel.<p>There are many sail technologies out there. A common question we get asked is “Are you going to use flettner rotors/suction airfoils?”. Both of these technologies use power supplied by the ship to increase the lift produced by a surface; rotor-sails spin, and suction airfoils…suck? Each of these have a place, especially at low vessel speeds. But our customers ask us for a solution that works for container ships cruising at the relatively high speed of 22kt. At these speeds, the relative wind is almost always ahead of you, so lift/drag becomes more important. Powered sails suffer from poor lift/drag, both from the high induced drag from very high lift coefficients, and system losses from drawing on ship’s power. So no we are not going with flettner rotors/suction airfoils. While they are the new exciting technology on the block, if you factor in their power usage and high drag ratio, they are just not as practical as a simple sail.<p>So now that we’ve given a general summary of sailing, it’s time to explain how a 747 wing will ever fit inside a 9ft tall cargo container. It’s simple really: imagine a tape measure. In a tape measure a thin, flexible strip of metal is wound into a spiral. Then, when the metal is uncoiled, it naturally returns to its original shape. That’s exactly how we plan to make our sails. The skin of our sail or the inner spars (we haven’t finalized our design) will be made of tape measure like material (2mm thick steel) and the wing will be able to extend out of the cargo container. The video in the first paragraph explains this in a bit more detail.<p>By fitting our sail into a cargo container we allow for our device to be installed on any cargo ship right at port. Remember how we mentioned that some shippers are ordering a lot more ships and some ships are getting retrofitted with new fuel? Well, shipyards are backed up for the next 5 years. By making a device that requires no shipyard to install, not only will we drastically outcompete other retrofit WASP companies in terms of deployment cost, but we will be the only company with a product shippers can put on their ship without a multiple year wait time.<p>Do you have any interesting stories around sailing or wind tech? We would love to hear your ideas, experiences, and feedback on any and all of the above!
Upvote: | 527 |
Title: This is the monthly "Co-Founder? Seeking Co-Founder?" post.<p>It would be great if whoishiring takes over this.<p>Example:<p>SEEKING CO-FOUNDER | FinTech | CEO/Sales | US/EU/Worldwide | Pre-Seed<p>A great idea in the taxation niche, MVP ready to go. Ideally someone with local connections to SMEs.
Upvote: | 47 |
Title: E.g., any special mouse, keyboard, ring, headset, ... that has made a significant impact on your workflow.
Upvote: | 377 |
Title: I am mainly familiar with AWS but the configuration of S3, lambdas (including creating layers for python dependencies), api gateway, cloudfront, and all of the obscure permissions are just a major pain in the butt. I spent maybe 5 hours building something in Jupyter and then an entire day just dealing with AWS.<p>Is there an easier way? If I gave you a python script and some html what would you use to get it behind a www in the least amount of time possible?
Upvote: | 82 |
Title: Hi HN,
I've been working on an experimental tool that helps you use GPT to work on your codebase. I'd love to improve the tool if there's interest. New ideas welcome! I think this could also be useful for experimenting with other types of recursive prompts.<p>It’s a little bit Swiss Army knife and a little bit skynet:<p><a href="https://github.com/ferrislucas/promptr">https://github.com/ferrislucas/promptr</a><p>From the README:
Promptr is a CLI tool for operating on your codebase using GPT. Promptr dynamically includes one or more files into your GPT prompts, and it can optionally parse and apply the changes that GPT suggests to your codebase. Several prompt templates are included for various purposes, and users can create their own templates.
Upvote: | 109 |
Title: Counting the comments in the who wants to be hired thread and dividing by the number of comments in who is hiring = 0.94.<p>A year ago it was 0.23. Here's a plot with the data from the past couple of years - https://pasteboard.co/tjZd0PgTI8GK.png<p>(posting this about a day after the April thread opened)
Upvote: | 363 |
Title: Hi HN, I’m Shimon, the co-founder of Datree: A policy management solution for Kubernetes. We help DevOps engineers prevent misconfigurations in their Kubernetes by enforcing an organizational policy on their clusters. Engineers can define a custom policy or use one of Datree’s built-in policies, such as NIST/NSA Hardening Guide, EKS Security Best Practices, CIS Benchmark, and more.<p>Our website is at https://datree.io and our GitHub is here: https://github.com/datreeio/datree<p>This is not the first time I have shown Datree to the HN community: A little over a year ago, I posted here an earlier version of Datree (https://news.ycombinator.com/item?id=28918850). At that time, Datree consisted of a CLI tool to detect Kubernetes misconfigurations during the development process (locally or in the CI/CD), unlike the version I present today in which the enforcement happens in production.<p>We built the CLI tool because we detected a big problem among Kubernetes operators: Misconfigurations. Kubernetes is extremely complex and flexible, which makes it very easy to poorly configure it in ways that are not secure. And indeed, we talked to dozens of Kubernetes operators who suffered from various problems, starting with failed audits, all the way to downtime in production, all because of misconfigurations.<p>Our solution was simple: Give the developers the means to shift-left security testing during the development process with a CLI tool that can be integrated into the CI/CD. We thought this was the best way to approach the problem: It is easiest to fix misconfigurations in the development process before they are deployed to production, it prevents context-switching and relieves resources from the DevOps team.<p>While the CLI tool was very popular among the open-source community (it got over 6000 stars on GitHub), we soon realized that CI/CD enforcement is not enough. As we talked with Datree’s users, we realized we had made a fundamental mistake: We thought of misconfiguration prevention in technical terms rather than organizational terms.<p>Indeed, from a technical point of view, it makes sense to shift-left Kubernetes security. But when considering the organizational structure in which it takes place, it simply isn’t enough. DevOps engineers told us that they love the shift-left concept, but they simply cannot rely on the goodwill of the engineers to run a CLI tool locally or to monitor all the pipelines leading to production. They need governance, something to help them stay in control of the state of their clusters.<p>Moreover, we realized that many companies who use Kubernetes are heavily regulated, and cannot take any chances with their security. Sure, these companies want the engineers to fix misconfigurations during development, but they also want something to make sure that no matter what, their clusters remain misconfiguration-free.<p>Based on this understanding, we developed a new version of Datree that sits on the cluster itself (rather than in the CI/CD) and protects the production environment by blocking misconfigured resources with an admission webhook. It has a centralized policy management solution to enable governance, and native monitoring to get real-time insights into the state of your Kubernetes.<p>I look forward to hearing your feedback and answering any questions you may have.
Upvote: | 43 |
Title: Hi HN,<p>I am studying robotics and exoskeletons as a hobby. Can HN please recommend high quality resources for self study?
Upvote: | 180 |
Title: C is a programming language and Unix is an OS. So, how did they run C before Unix era?
Upvote: | 53 |
Title: ChatIDE is a sort of a DIY Github copilot X.<p>Launch ChatIDE with a `Cmd + Shift + i`. You bring your own OpenAI api key. You can set the system prompt, temperature, and max tokens in the extension settings.<p>The extension is currently not aware of any of the files in your project, although that might be cool to add later. Use it just like you would use ChatGPT in the browser. Feedback and contributions more than welcome.
Upvote: | 43 |
Title: Hi HN, we're Stew and Stefan from Type (<a href="https://type.ai/">https://type.ai/</a>). We're building an AI-first document editor that helps you write. It's similar to Notion, but focused on building a solid authoring experience.<p>Here’s a general demo: <a href="https://www.youtube.com/watch?v=MpK9PWo0lUw">https://www.youtube.com/watch?v=MpK9PWo0lUw</a><p>And here’s a demo that includes math and code blocks: <a href="https://type.ai/code-math-demos">https://type.ai/code-math-demos</a><p>There are a lot of AI writing products out now, but we've found that most of them treat writing like a one-shot activity that should be delegated to AI. We don't think that's the optimal way to write. We think of great writing as the product of clear thinking, which requires a lot of time tinkering with and refining ideas. So we’re building a user-friendly document editor that puts the author front and center.<p>As you write in Type, you can press cmd+k to summon simple AI commands. Most of our commands are grounded in familiar writing primitives (ex. “Write paragraph”) and attempt to understand the context of your document.<p>Type supports multiple rich block types, including code and math and our commands are able to both interpret and output these block types. So if you're writing an introductory essay about machine learning, for example, you can use Type's chat feature to generate and refine equations and code blocks you'd like to include in your document. Once you’re satisfied with what Type has generated, you can drag and insert the block anywhere in your document (as seen in the demo video above).<p>We’ve also built a "what to write about next" feature in the document sidebar that offers suggestions on ideas you may consider adding to your document.<p>We’ve built some editor features that aren’t AI-specific but which we think make for an enjoyable authoring experience: (1) Type is built from the ground up to be offline-first. This means most interactions (search, loading documents, etc.) are instant; (2) Mobile support as installable PWA; (3) Keyboard shortcuts for the most useful commands; (4) Markdown copy/paste support.<p>We designed Type to be most useful for longer-form writing, so we encourage you to try it out in the context of something like an essay or a technical tutorial. If you try it out at <a href="https://type.ai">https://type.ai</a>, we’d love to hear what you think. We think Type feels pretty different from other AI writing tools that produce fairly shallow content, but would love to get your honest feedback on whether we're hitting the mark.<p>Each account comes with a free allocation of AI commands, after which you can activate a paid plan for unlimited AI usage (you can still create and access unlimited docs on the free plan). If you'd like some additional free credits, please just drop us a note at [email protected] and we'll refill your free credits.<p>We'd love your feedback on what feels helpful and what feels confusing or missing. Thanks!
Upvote: | 194 |
Title: Have you been doing this for a long time? Do you change your schedule for the weekends?<p>For me its easy for me to avoid eating during the week, much more than controlling my eating habits and eating like an adult... but then weekend comes and it becomes 10X harder, especially if others are eating around me. And then a sunny day comes and you "have" go have some barbecue and beer, an hamburguer at night, etc.
Upvote: | 51 |
Title: We are being dizzyingly bombarded with submitted content and information about GPT and AI by the hour, and one way to filter this excess is to create a tab only on the subject as exists with Show and Jobs.
Upvote: | 77 |
Title: Hey HN, we’re Daniel and Oscar, founders of GradientJ (<a href="https://gradientj.com">https://gradientj.com</a>), a web application that helps teams develop, test, and monitor natural language processing (NLP) applications using large language models (LLMs).<p>Before GradientJ, we’d been building NLP applications for 4 years, using transformer models like BERT. With the advent of LLMs and their zero-shot/few-shot capabilities, we saw the NLP dev cycle get flipped on its head. Rather than having to hire an army of data labelers and data scientists to fine-tune a BERT model for your use case, engineers can now use LLMs, like GPT-4, to build NLP endpoints in minutes.<p>As powerful as this is, the problem becomes that without appropriate tools for version control, regression testing, and ongoing maintenance like monitoring and A/B testing, managing these models is a pain. Because the data being evaluated is often fuzzy, developers either have to build complex text processing regex pipelines or manually evaluate each output before a new release. Moreover, if your prompts are only maintained in a notion doc or google sheet, completely separate from these tests, it’s difficult to identify what the changes were that led to underperformance. The workflow often devolves into manual and subjective human data labeling just to decide if new versions of your model are “good enough” to deploy.<p>GradientJ is a web application and API to address that. We let you iterate on prompts, automatically regression test them along multiple dimensions, and finally manage them once deployed.<p>You’d think these are pretty straightforward things to build, but we’ve noticed most versions of “LLM management apps” focus on organizing the workflow for these components without dramatically improving on automating them. At the end of the day, you still have to pass your side-by-side prompt comparison through the “eye-ball test” which creates processes that are bottlenecked by human time. We think by using the very same technology, NLP, you can dramatically reduce the developer labor required for each of these steps.<p>Here’s how we do it:<p>For prompt iteration, rather than just a text-editor “playground” with some special syntax to delineate variables, we’re trying to use large language models to create a Copilot-like experience for prompt engineering. This means aggregating all the tricks of prompt engineering behind a smart LLM assistant who can suggest ways to restructure your prompt for better output. For example, when someone just wants their output in JSON form, we know where to inject the appropriate text to nudge the model towards generating JSON. When combined with our regression testing API, those prompt suggestions will actually be based on the specific dimensions of prompt underperformance. The idea is that the changes required to make a prompt’s output follow a certain structure are different from the ones you’d make to have the output follow a certain tone.<p>When it comes to testing, even before LLMs, configuring high quality tests for expressive NLP models has historically been hard. To compare anything more complicated than classification labels, most people resort to raw fuzzy string comparisons, or token distribution differences between the output. We’re trying to make automated NLP testing more objective by using LLMs to actually power our regression testing API. We use NLP models to provide comparisons between text outputs along custom dimensions like “structure”, “semantics”, and “tone”. This means before you deploy the latest version of your email generation model, you know where it stands along each of the discrete dimensions you care about. Additionally, this helps prevent your prompt engineering from becoming a game of “whack-a-mole”:overfitting your prompt on the handful of examples you can copy and paste while developing.<p>For deployment, we provide a stable API that always goes to the latest iteration of a prompt you’ve chosen to deploy. This means you can push updates over-the-air without having to change the API code. At the same time, we’re tracking the versions used for inference under the hood. This lets you use that data to further improve your regression tests, experiment with fine-tuning across other providers or open source models, or set up alerts around prompt performance.<p>Each of these pieces of our product can be used in isolation or all together, depending on what the rest of your NLP infrastructure looks like.<p>If you use LLMs and are looking for ways to improve your workflow, or if you need to build NLP applications fast and want to bypass the traditional slow data labeling process, we’d love your feedback!
Upvote: | 56 |
Title: Want something better than k-means? I'm happy to announce our SOTA k-medoids algorithm from NeurIPS 2020, BanditPAM, is now publicly available! `pip install banditpam` or `install.packages("banditpam")` and you're good to go!<p>k-means is one of the most widely-used algorithms to cluster data. However, it has several limitations: a) it requires the use of L2 distance for efficient clustering, which also b) restricts the data you're clustering to be vectors, and c) doesn't require the means to be datapoints in the dataset.<p>Unlike in k-means, the k-medoids problem requires cluster centers to be actual datapoints, which permits greater interpretability of your cluster centers. k-medoids also works better with arbitrary distance metrics, so your clustering can be more robust to outliers if you're using metrics like L1. Despite these advantages, most people don't use k-medoids because prior algorithms were too slow.<p>In our NeurIPS 2020 paper, BanditPAM, we sped up the best known algorithm from O(n^2) to O(nlogn) by using techniques from multi-armed bandits. We were inspired by prior research that demonstrated many algorithms can be sped up by sampling the data intelligently, instead of performing exhaustive computations.<p>We've released our implementation, which is pip- and CRAN-installable. It's written in C++ for speed, but callable from Python and R. It also supports parallelization and intelligent caching at no extra complexity to end users. Its interface also matches the sklearn.cluster.KMeans interface, so minimal changes are necessary to existing code.<p>PyPI: <a href="https://pypi.org/project/banditpam" rel="nofollow">https://pypi.org/project/banditpam</a><p>CRAN: <a href="https://cran.r-project.org/web/packages/banditpam/index.html" rel="nofollow">https://cran.r-project.org/web/packages/banditpam/index.html</a><p>Repo: <a href="https://github.com/motiwari/BanditPAM">https://github.com/motiwari/BanditPAM</a><p>Paper: <a href="https://arxiv.org/abs/2006.06856" rel="nofollow">https://arxiv.org/abs/2006.06856</a><p>If you find our work valuable, please consider starring the repo or citing our work. These help us continue development on this project.<p>I'm Mo Tiwari (motiwari.com), a PhD student in Computer Science at Stanford University. A special thanks to my collaborators on this project, Martin Jinye Zhang, James Mayclin, Sebastian Thrun, Chris Piech, and Ilan Shomorony, as well as the author of the R package, Balasubramanian Narasimhan.<p>(This is my first time posting on HN; I've read the FAQ before posting, but please let me know if I broke any rules)
Upvote: | 281 |
Title: I get that YouTube and virtually all social media sites want to maximize clicks. I get that YouTube really doesn't give a rat's ass about what I want in that context: they think it's worth far more to override my preferences and add all kinds of channels to my feed than I want. But if you have direct knowledge of the algorithm, please tell me why I can watch a video to the end, then have it recommended for days, weeks, or months after, even when I have dutifully upvoted or commented. What's in it for YouTube? Just because I liked a video by Kyle Kulinsky or Jordan Peterson in 2022 doesn't mean I'm still dying to watch it in 2023. And if you want to force shit down my throat, why give up valuable screen real estate to videos I already saw?
Upvote: | 64 |
Title: Posting from a throwaway account to maintain privacy.<p>This project is a salvo against leetcode-style interviews that require candidates to study useless topics and confidently write code in front of a live audience, in order to get a job where none of that stuff matters.<p>Cheetah is an AI-powered macOS app designed to assist users during remote software engineering interviews by providing real-time, discreet coaching and integration with CoderPad. It uses Whisper for audio transcription and GPT-4 to generate hints/answers. The UI is intentionally minimal to allow for discreet use during a video call.<p>It was fun dipping into the world of LLMs, prompt chaining, etc. I didn't find a Swift wrapper for whisper.cpp, so in the repo there's also a barebones Swift framework that wraps whisper.cpp and is designed for real-time transcription on M1/M2.<p>I'll be around if anyone has questions or comments!
Upvote: | 301 |
Title: Well this was a major surprise so I figured I’d share it here to get some eyeballs on it.<p>Essentially, the latest iOS (16.4 at post time) allows your cellular carrier (via eSIM) to add “managed networks” to your device.<p>These networks cannot be removed, they cannot have “automatically join” disabled, and they have equal priority with your real, personal networks.<p>So guess what happens when your neighbors get a wifi/modem combo that blasts a free hotspot SSID? Not only does it pollute the already crowded 2.4ghz band, your iPhone will often prefer this connection over your real /local wifi (despite said wifi being at 1 bar).<p>As of post-time, there is no way to remove these networks short of completely disabling cell service/removing the eSIM and resetting all network settings.<p>You can see this for yourself by going to WiFi/“edit” and scrolling down.<p>Edit: to clarify, I can disable “auto join”, but in 4-5 minutes all of my devices have auto-join turned back on. I’m guessing it re-syncs with the carrier profile. Also, this does not seem to be eSIM or SIM related it can happen on both.
Upvote: | 1025 |
Title: Hi, I am David Kircos. The Founder of Quadratic (<a href="https://QuadraticHQ.com" rel="nofollow">https://QuadraticHQ.com</a>), an open-source spreadsheet application that supports Python, SQL (coming soon), AI Prompts, and classic Formulas.<p>Unlike other spreadsheets, Quadratic has an infinite canvas (like Figma). As a result, you can pinch and zoom to navigate large data sets, and everything renders smoothly at 60fps.<p>Our vision is to build a place where your team can collaborate on data analysis. You can write Python, AI Prompts, and Formulas in one spreadsheet feeding each other data and updating automatically.<p>Quadratic is built using WebGL and Rust WASM. To render a large grid of cells smoothly, we tile the spreadsheet similar to google maps. If you are interested in the technical details, check us out on GitHub (<a href="https://github.com/quadratichq/quadratic/">https://github.com/quadratichq/quadratic/</a>)<p>You can use AI to help you write Python and then run the code directly in Quadratic. Then, we feed the result back to the AI model so it can follow along, help you debug, and modify your existing code.<p>AI can also be used to directly generate data onto the sheet with prompts. It knows the context of what's on the sheet and how the data it's inserting fits in. Try it out.<p>SQL is coming soon... stay tuned!
Upvote: | 224 |
Title: Other than treating what is possibly ADD, what has helped you focus and just get your fingers on the keyboard?<p>Edit: perhaps I should add, I’m not really distracted by exernal things but perhaps the complexity of the problem / poorly engineered code just adds so much inertia to getting the fingers typing away. I’m a prototyping research scientist so producing well written code is not my strong suit, it’s more like let’s build something new and see if we can get it to work.
Upvote: | 90 |
Title: I want to draw attention to Bob Lee, a well-respected technologist and prototype hacker, always curious and sharing lots of interesting technical developments. He was a great role model for how Engineers should be respected in an executive capacity as he advanced his career from 'Software' to 'Product'. His efforts contributed to technology used by millions. What happened to him is tragic and wrong; he deserved better. Thank you, I'll miss you 'crazybob'.<p>Please share your stories featuring Bob Lee, who I'm sure would like to be remembered for his contributions rather than as a victim of this unfortunate awful event.
Upvote: | 2090 |
Title: Lately I've felt exhausted due to the deluge of AI/GPT posts on hacker news, and have seen similar grumblings. I threw together this frontend that filters out anything with the phrases AI, LLM, GPT, or LLaMa for use until the hype dies down a bit.<p>Before anyone asks, yes I did try to use ChatGPT to help, and while the code it provided was helpful, it needed some heavy bug-fixing.<p>Edit: One other note I forgot to mention. The favicon is generated by Stable Diffusion, I asked it to generate an "Aritificial Intelligence Favicon", and then I added the red circle with line through it.
Upvote: | 69 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.