A mysterious chatbot called “gpt2-chatbot” has sparked much excitement and speculation within the AI community.
This chatbot that you may find and check out for yourself LMSys Chatbot Arenais touted by some users as more advanced than current state-of-the-art models akin to GPT-4 and Claude 3 Opus.
Seemingly created out of nowhere, its origins are completely unknown. Could it’s an experimental project from a number one AI lab or a wierd leak?
There is a mysterious recent model called gpt2-chatbot that’s accessible through a serious LLM benchmarking site. Nobody knows who made it or what it’s, but I've played with it a bit and it appears to be concerning the same performance level as GPT-4. A mysterious GPT-4 class model? Clean! pic.twitter.com/1s2iEreaiT
– Ethan Mollick (@emollick) April 29, 2024
Reddit users mainly think it comes from OpenAI because the model admits that the response style and tone seem like different than GPT-4 and GPT-4 Turbo, however the knowledge threshold is different from that currently available OpenAI Models.
gpt2-chatbot is a madman on the subject of ASCII art and is miles ahead of every other model pic.twitter.com/1IwEaJyU7Q
— Phil (@philll__1) April 29, 2024
I asked gpt2-chatbot myself: “Who are you?”, to which he replied:
When I asked him why it was called gpt2-chatbot, he replied:
The performance of gpt2-chatbot sparks speculation
Regarding performance, a user is on the Y-Combinator forum discussed the bot's expertise in materials science and discussed the creation of turbine blades with a depth they’d not have otherwise experienced.
Another user, well-known data scientist Simon Wilson, was impressed by its handling of highly personalized queries, suggesting it has an in depth understanding of individuals's personal backgrounds.
You can try the mysterious gpt2 chatbot at https://t.co/4ARf9G6RHb (select “Direct Chat” and choose it from the menu)
First impression: I’m very impressed. It gave me a greater answer to a first-person search (“Who is Simon Willison?”) than every other model I’ve tried https://t.co/MCJMVhNs8k
— Simon Willison (@simonw) April 29, 2024
Others identified standard LLM deficiencies akin to hallucinations and factual inaccuracies.
Having used it myself, I’d agree that it looks like a variant of the GPT-4.
What some are speculating about whether it is healthier than GPT-4 is to spotlight that many felt that GPT-4's performance declined over time (which could have coincided with). OpenAI Provision of computing resources for training recent models).
So this potentially unadulterated variant might feel like GPT -4 did when it first got here out. However, should you add a slight differentiation in behavior and performance on specific tasks, you may see why the speculation is so high.
When discussions about gpt2-chatbot However, most will not be sure whether it represents a concrete advance over current models.
Overall, there isn’t any way that an AI company could train a wholly recent chatbot in this fashion and release it in this fashion. It is sort of definitely a GPT-4 variant.
Ultimately, it stays just hearsay until the creator comes out from behind the scenes and claims ownership.
