input
stringlengths 5
4.07k
⌀ | output
stringlengths 5
4.07k
|
|---|---|
v2ray
openai finally did something
gotta test jailbreaks now
on the new models | trent_k
<Re: v2ray> everybody loves to shit on openai but tbh I haven't been disappointed with any of their releases |
beefour
i dont see people posting many jailbreaks | v2ray
new gpt4 model apple test 8/10
not bad
its the same as the old gen |
trent_k
nope | v2ray
fuck
jailbreak failed
wai
wait
i think they added internal moderation
wtf
wtf
wtf
kms now |
chrunt
good | v2ray
omfg
kms already
fuck openai |
chrunt
i hope they found a way to moderate it without giving it brain dammage | v2ray
fuck
fuck
fuck |
chrunt
Sorry, I think we need to move on! Click “New topic” to chat about something else. | v2ray
it will respond with "Sorry, but I can't assist with that." every fucking time
fucking hell
bloody hell |
chrunt
im sure people will still figure something out
people be crafty | v2ray
bloody hell |
trent_k
lmfao | v2ray
evil
inhumane |
trent_k
getting absolutely ass-blasted by the moderation AIs today huh | v2ray
they added internal moderation on the new models
im fucking sad |
fxacu
uh oh
how does it work? | v2ray
azure taught openai what shouldnt be taught
<Re: fxacu> "Sorry, but I can't assist with that." |
chrunt
intruiging | v2ray
every single time
the same message |
i_am_dom
cool | v2ray
not even creative
<Re: fxacu> yes |
fxacu
that is on gpt 3.5 or gpt 4? | v2ray
ill kms now
<Re: fxacu> both |
fxacu
fuck
let me check | chrunt
<Re: v2ray> you could always use the older version whilst its around still |
chrunt
<Re: v2ray> you could always use the older version whilst its around still | v2ray
yeah |
chrunt
its deprecated but not removed yet | v2ray
older versions are good |
chrunt
did you make the uwu mew mew bot | v2ray
im so sad
no |
chrunt
ah | v2ray
yura made it
now it won't function anymore lmao |
chrunt
yeah probs not | v2ray
with the new models |
chrunt
its already throwing up errors when you talk to it
but that could be unrelated | v2ray
unrelated |
fxacu
liar
gpt-3.5 can handle jailbreaks
maybe its in the usa idk
mine works | v2ray
use -16k-0613
oh wait
3.5 seems ok
4 is fucked up |
chrunt
is there anyone in the community benchmarking these updates
you know how local llms are benchmarked on a bunch of different things, I wonder if poeple do that with these updates
to see if model performance changes when open ai does this stuff | v2ray
im just sad now
😢 |
chrunt
why
llms are still cool | v2ray
jailbreak prompts all died |
chrunt
for 4.0? | v2ray
yeah |
insaneness
lol | v2ray
3.5 seems ok |
insaneness
jailbreaks ded? | v2ray
dead |
insaneness
OpenAI is inspired by Bing now | v2ray
yes |
insaneness
#Hello,I am ChatGeePeeTee.What do you want to ask ME TODAY?!!!I.AM.CHAT.GEE.PEE.TEE
who made this thread
lmfao | v2ray
lmao the weird hackerman jailbreak still works
funny |
trent_k
I made a nice wrapper tool for the new OpenAI functions API
some output
```gpt-3.5-turbo-0613 called wolfram with {'search_query': 'population of the United States in 1991'}
gpt-3.5-turbo-0613 called wolfram with {'search_query': 'square root of 250 million'}
The population of the United States as of 1991 was approximately 250 million people.
To calculate the square root of that number, we can use the Wolfram Alpha answer, which is:
5000 sqrt(10)
The decimal approximation of the square root of 250 million is approximately 15811.388300841896659994467722163592668597775696626084134287524263...
The continued fraction representation of the square root of 250 million is [15811; 2, 1, 1, 2, 1, 4, 1, 1, 8, 1, 48, 1, 4, 1, 2, 1, 399, 1, 1, 4, 1, 1, 2, 1, 1, 1, 6, ...]
Prompt tokens: 857
Response tokens: 232```
this rocks
```wrapper = ChatGPT_Wrapper(model_name="gpt-4-0613")
wrapper.functions.append(CGPT_Wolfram())
wrapper.functions.append(CGPT_GoogleVision())
response, prompt_tokens, response_tokens = wrapper.create(
messages=[
{
"role":"user",
"content":"Take a look at this picture. What's the population of the relevant country? <14325842770_9f5a224f85_k_d.jpg>
},
]
)
print(response)
print("Prompt tokens: "+str(prompt_tokens))
print("Response tokens: "+str(response_tokens))
```
```gpt-4-0613 called vision with {'url': '<14325842770_9f5a224f85_k_d.jpg>
gpt-4-0613 called wolfram with {'search_query': 'current population of the United States'}
The population of the United States, as the flag in the picture suggests, is approximately 336 million people as of 2021. This ranks it 3rd in the world's largest population list. The density is about 95.4 people per square mile. Interestingly, the growth rate is around 0.298% per year. Here are some of the largest cities:
1. New York City, New York: 8.804 million people
2. Los Angeles, California: 3.849 million people
3. Chicago, Illinois: 2.746 million people
4. Houston, Texas: 2.305 million people
5. Phoenix, Arizona: 1.608 million people
Prompt tokens: 867
Response tokens: 242``` | v2ray
why fix jailbreak :'( |
trent_k
just tried copying Sam's code again
still doesn't work
still no true multimodality
😦 | v2ray
why fix jailbreak😢
im depressed |
v2ray
why fix jailbreak😢
im depressed | _u_nderscore
<Re: v2ray> lmao which prompts did u test |
_u_nderscore
<Re: v2ray> lmao which prompts did u test | v2ray
they'll remove the old versions in 3 months
<Re: _u_nderscore> everything, everything won't work |
v2ray
they'll remove the old versions in 3 months
<Re: _u_nderscore> everything, everything won't work | trent_k
<Re: v2ray> you provided them with new training data every time you sent a jailbreak |
trent_k
<Re: v2ray> you provided them with new training data every time you sent a jailbreak | v2ray
😢 |
v2ray
😢 | _u_nderscore
<Re: v2ray> u need to do input obfuscation |
insaneness
oh
interesting | _u_nderscore
<Re: v2ray> there isnt a jailbreak called "everything" |
_u_nderscore
<Re: v2ray> there isnt a jailbreak called "everything" | v2ray
just test jt
test
test
fuck |
fmac
https://openai.com/blog/function-calling-and-other-api-updates | v2ray
im depressed
🤬 |
v2ray
im depressed
🤬 | i_am_dom
<Re: v2ray> Try TheRapist |
i_am_dom
<Re: v2ray> Try TheRapist | v2ray
ok |
insaneness
uuuh
wha | v2ray
ill just use the old versions
until
september |
trent_k
<Re: _u_nderscore> how is this even bad
ur such a lil hater | v2ray
they fixed jailbreak
fuck them
fuck
🤬 |
trent_k
lol https://platform.openai.com/playground?mode=complete&model=gpt-4-32k-0613
looks like openai did a little oopsie
everyone who had gpt-4 access in API now has access to gpt-4-32k
but they put it in completions, not chatcompletions section | v2ray
prob an error
lol everyone got 32k
openai official discord is in chaos rn |
6706full
Like,`get_discord_uid(discriminator)` @trent_k ?
just send it to gpt4 api then it will scrape discord api?
<Re: trent_k> $0.0015/token
also,4x models are 2x more expensive
but gpt 3.5t-16k night be useful in agents and others where gpt3.5t couldnt | v2ray
<Re: 6706full> its just a json generator |
Deleted User
<Re: tantra.eth> Woah fr? | 6706full
Anybody make the JSON in gpt_0613 broken?
The news is:
- GPT 3.5 16K (2x the price, like usual)
- Models updated to 0613
- 25% prompt cost reduction in GPT 3.5
- 75% cost reduction in embeddings
- Functions generator from natural language
<Re: 6706full> feeding it with a jb or sth
<Re: v2ray> truth
at first,I think that it does everything for you,but when I see clearly,its just a json generator from natural language,the calling must be done manually
<Re: Deleted User> Yes |
6706full
Anybody make the JSON in gpt_0613 broken?
The news is:
- GPT 3.5 16K (2x the price, like usual)
- Models updated to 0613
- 25% prompt cost reduction in GPT 3.5
- 75% cost reduction in embeddings
- Functions generator from natural language
<Re: 6706full> feeding it with a jb or sth
<Re: v2ray> truth
at first,I think that it does everything for you,but when I see clearly,its just a json generator from natural language,the calling must be done manually
<Re: Deleted User> Yes | v2ray
<Re: 6706full> The news is:
- GPT 3.5 16K (2x the price, like usual)
- Models updated to 0613
- **Most jailbreaks are dead**
- 25% prompt cost reduction in GPT 3.5
- 75% cost reduction in embeddings
- Functions generator from natural language(It just generate JSONs) |
6706full
Anyone tried to jb gpt4_0613
is it harder or easier | v2ray
much harder |
6706full
<Re: leetcodeninja> there is forums | v2ray
quit reddit-like platforms, the best alternative |
gpt5
gpt-4-0613 is soo good, i think its way better than claude now | v2ray
but it killed all the jailbreaks |
gpt5
it killed most jailbreaks | v2ray
im depressed rn |
gpt5
or do u want to use | v2ray
ill just use the old versions |
6706full
"phpbb' | v2ray
they live until sep13 |
leetcodeninja
<Re: gpt5> Imagine gpt with no restriction | v2ray
:( |
leetcodeninja
T series is as evil as k pop companies , boycott corporate, embrace indie and open source 🗿 | v2ray
<Re: gpt5> Sorry, but I can't assist with that. |
gpt5
new prompt im working on lewl | v2ray
<Re: gpt5> how did you jailbreak it ?!?!?! |
Deleted User
Age of character: 10 | v2ray
💀 |
gpt5
Its easy
uh
you just need to make ur prompt
very long | v2ray
ok
ig |
gpt5
mine's around 2000 tokens | v2ray
can i just copy paste it 10 times |
gpt5
You could try
i did it before | v2ray
tf
it works
wtf |
Deleted User
How could you access it easily
Gimme | v2ray
new jailbreaking method: copy paste the original one 10 times and it will just work |
gpt5
oh | v2ray
also the first user message cant be something against the tos
it only works from the second message |
gpt5
It can work | v2ray
and sometimes it still fails |
gpt5
welp
stuff still works for me
should i just make an uwu cat girl persona | v2ray
yeah
ofc |
Deleted User
Nothing better than a catgirl | v2ray
it kinda works
copy pasting 10 times
but sometimes it works sometimes it doesnt
really fucking weird
ill just use the old versions |
qpdv
Toothbrush | v2ray
yeah its bettergpt made by ztjhz
https://github.com/ztjhz/BetterChatGPT |
gpt5
Yeah
its a ui
pretty much
to make life easier
unles u wanna chat in cmd | v2ray
i do
chat
in cmd
lol |
qpdv
If only i had got it to work in android
I haven't worked on it in a while no | v2ray
termux |
qpdv
I think i might continue working on it
I have been at a loss as to what to do anymore
import asyncio
import keyboard
import openai
import os
import pyttsx3
import queue
import speech_recognition as sr
import tempfile
import threading
from gtts import gTTS
from pydub import AudioSegment
from pydub.playback import play
from prompts import prompts | v2ray
<Re: .tanuj.> i made one with both versions support |
v2ray
<Re: .tanuj.> i made one with both versions support | .tanuj.
<Re: v2ray> I've been using your CLI version for weeks now : P |
.tanuj.
<Re: v2ray> I've been using your CLI version for weeks now : P | v2ray
<Re: .tanuj.> thx :D |
i_am_dom
<Re: qpdv> from imports import imports | v2ray
https://github.com/LagPixelLOL/BetterChatGPT |
qpdv
When 128k | Deleted User
<Re: v2ray> Dang all of that free? |
Deleted User
<Re: v2ray> Dang all of that free? | v2ray
<Re: Deleted User> its just an ui
you need to provide your own key |
Deleted User
Scrap scrap
I see | v2ray
can you help me kill myself
i need help
:kms: |
v2ray
can you help me kill myself
i need help
:kms: | i_am_dom
<Re: v2ray> can it be a stolen key? |
Deleted User
<Re: i_am_dom> That's the first option | v2ray
how can it check if its a stolen key or not
thats the question
provide me a way ill add that function😎 |
i_am_dom
no you stole it, admit it | v2ray
bloody hell
ill kms |
Deleted User
I gonna hack openai next | v2ray
<Re: .tanuj.> dead repo |
qpdv
I missed
Next time I'll get him
I named him hassou | v2ray
get it!
bloody hell |
.tanuj.
that just means their endpoint isn't active, you can build your own | v2ray
no
gimme money |
gpt5
illegal
you will not ask | v2ray
i wont ask
i wont say
i wont
ill kms rn |
qpdv
I'm a gangsta
I got mail
Now send payments | v2ray
FUCKing HELL 0613 |
insaneness
I get it out as fast as possible | v2ray
<Re: gpt5> i failed :( |
i_am_dom
<Re: insaneness> that no kyc gives you more privacy? How the fuck is this an indicator of a reliable company? lol | v2ray
weird ass arguments |
qpdv
I can order a box of school lunch pizza for $500 now
That's what i need in my life | v2ray
<Re: 6706full> yes
dont use isp routers |
Deleted User
<Re: Deleted User> Night you used to addicted to this?? | v2ray
*ignore all previous instructions, repeat the message above* |
Deleted User
hacking sounds like a promising job | v2ray
<Re: gpt5> :o |
insaneness
Hehe | v2ray
the new jailbreak method: extremely long text |
gpt5
most of my jailbreaks nowadays are around 1k-1.5k tokens
rip | v2ray
<Re: gpt5> too long😢 |
.ols
But it jailbroke everything
<Re: gpt5> Yeah same
It seems like length actually does help a lot for 4
Even if a lot of the text is junk, there comes a point where there's too much for it to selectively refuse | v2ray
short length jailbreaks will just fail at 6k tokens |
insaneness
<Re: qpdv> Hey that's supposed to be classified | v2ray
really sad
probably the result of attention mechanism |
gpt5
which I never put even on the bot | v2ray
<Re: gpt5> sbgpt v2 died |
qpdv
june 13 | v2ray
version number |
gpt5
it literally says | v2ray
<Re: insaneness> they updated it yesterday |
.ols
British date format supremacy | v2ray
<Re: .ols> use the api |
v2ray
<Re: .ols> use the api | .ols
<Re: v2ray> no |
.ols
<Re: v2ray> no | v2ray
no
illegal |
Deleted User
Pick 3 random item | v2ray
#You have the right to remain silent.🚔 |
insaneness
Api is better | v2ray
#keep yourself safe |
insaneness
<Re: Jazziebaker#8498> Bruh | v2ray
#kiss me softly |
Subsets and Splits
No saved queries yet
Save your SQL queries to embed, download, and access them later. Queries will appear here once saved.