Hugging Face's logo
Nicolay Rusnachenko's picture
7 24 41

Nicolay Rusnachenko

nicolay-r
ritvik77's profile picture aldo2467's profile picture pennpenn23's profile picture
·
https://nicolayr.com/

AI & ML interests

NLP for Healthcare ⚕️ @BU_Research・PhD in NLP / IR ・Textual Information Retrieval

Recent Activity

reacted to hesamation's post with 👀 about 10 hours ago
longer context doesn't generate better responses. it can even hurt your llm/agent. 1M context window doesn't automatically make models smarter as it's not about the size; it's how you use it. here are 4 types of context failure and why each one happens: 1. context poisoning: if hallucination finds its way into your context, the agent will rely on that false information to make its future moves. for example if the agent hallucinates about the "task description", all of its planning to solve the task would also be corrupt. 2. context distraction: when the context becomes too bloated, the model focuses too much on it rather than come up with novel ideas or to follow what it has learned during training. as Gemini 2.5 Pro technical report points out, as context grows significantly from 100K tokens, "the agent showed a tendency toward favoring repeating actions from its vast history rather than synthesizing novel plans". 3. context confusion: everyone lost it when MCPs became popular, it seemed like AGI was achieved. I suspected there is something wrong and there was: it's not just about providing tools, bloating the context with tool use derails the model from selecting the right one! even if you can fit all your tool metadata in the context, as their number grows, the model gets confused over which one to pick. 4. Context Clash: if you exchange conversation with a model step by step and provide information as you go along, chances are you get worse performance rather than providing all the useful information at once. one the model's context fills with wrong information, it's more difficult to guide it to embrace the right info. agents pull information from tools, documents, user queries, etc. and there is a chance that some of these information contradict each other, and it's not good new for agentic applications. check this article by Drew Breunig for deeper read: https://www.dbreunig.com/2025/06/26/how-to-fix-your-context.html?ref=blog.langchain.com
replied to hesamation's post about 10 hours ago
longer context doesn't generate better responses. it can even hurt your llm/agent. 1M context window doesn't automatically make models smarter as it's not about the size; it's how you use it. here are 4 types of context failure and why each one happens: 1. context poisoning: if hallucination finds its way into your context, the agent will rely on that false information to make its future moves. for example if the agent hallucinates about the "task description", all of its planning to solve the task would also be corrupt. 2. context distraction: when the context becomes too bloated, the model focuses too much on it rather than come up with novel ideas or to follow what it has learned during training. as Gemini 2.5 Pro technical report points out, as context grows significantly from 100K tokens, "the agent showed a tendency toward favoring repeating actions from its vast history rather than synthesizing novel plans". 3. context confusion: everyone lost it when MCPs became popular, it seemed like AGI was achieved. I suspected there is something wrong and there was: it's not just about providing tools, bloating the context with tool use derails the model from selecting the right one! even if you can fit all your tool metadata in the context, as their number grows, the model gets confused over which one to pick. 4. Context Clash: if you exchange conversation with a model step by step and provide information as you go along, chances are you get worse performance rather than providing all the useful information at once. one the model's context fills with wrong information, it's more difficult to guide it to embrace the right info. agents pull information from tools, documents, user queries, etc. and there is a chance that some of these information contradict each other, and it's not good new for agentic applications. check this article by Drew Breunig for deeper read: https://www.dbreunig.com/2025/06/26/how-to-fix-your-context.html?ref=blog.langchain.com
reacted to kanaria007's post with 👀 about 10 hours ago
✅ New Article on Hugging Face: Structured Perception for Structured Reasoning — Rethinking AI Input Through the Five-Sense Protocol Title: 🧠 Understanding the Perceptual-Interface Protocol: Structured Sensory Modules for Cognitive Input Parsing 🔗 Read it here: https://huggingface.co/blog/kanaria007/understanding-the-perceptual-interface-protocol --- Summary: What if artificial intelligence systems could “sense” inputs — not physically, but structurally? This article introduces the *Five Sense Protocol*, a *theoretical blueprint* for embedding cognitive input structuring into AI systems. Inspired by human perceptual organization, it proposes *five abstract sensory layers* to refine reasoning through structured perception. --- Why It Matters: Current LLMs treat input as undifferentiated text streams. But humans don’t. We segment, anticipate, detect contradiction, monitor coherence, and adjust ethically — *before* reasoning begins. The Perceptual-Interface protocol brings this pre-reasoning perceptual organization into AI cognition design. --- Core Layers: • Surface Syntax Detection — using formatting as reasoning cues • Temporal Rhythm Tracking — anticipating structural pacing • Conflict Sensitivity — flagging contradictions and dissonance • Structural Coherence Mapping — detecting fragmented logic • Ethical Context Filters — triggering meta-awareness in sensitive domains --- 🧩 *Think of it as structured input for structured output.* Reasoning is only as good as the way it begins. --- Relevant For: • AI researchers exploring perceptual architecture in cognition • Developers building input-sensitive autonomous systems • Cognitive scientists bridging human and artificial attention models --- 🧠 Protocol Dataset: https://huggingface.co/datasets/kanaria007/agi-structural-intelligence-protocols --- *This isn’t perception emulation.* It’s *perception structuring* — for the next layer of intelligent reasoning.
View all activity

Organizations

None yet