The mysterious AI called “gpt2-chatbot” is back as "im-a-good-...

Capabilities seem to exceed GPT-4, Gemini 1.5, Claude, and anything else currently available.
Here's EVERYTHING you need to know (and how to try it for free):
Take the speculation with a grain of salt, and try the model for yourself to see the difference.
As more details come out, I'll add to this thread.
Bookmark and follow me @rowancheung to stay updated.
Last week, a new model called 'gpt-2 chatbot' appeared on the chatbot arena.
It was removed, but last night two new models called 'im-a-good-gpt2-chatbot' and 'im-also-a-good-gpt2-chatbot' appeared that seem to be new versions
therundown.ai/p/gpt2-mystery…
Before 'im-a-good-gpt2-chatbot' and 'im-also-a-good-gpt2-chatbot' were released, Sam Altman tweeted it.
OpenAI is not openly claiming that the new models are built by them, but this makes it seem pretty obvious that it is.

The model im-also-a-good-gpt2-chatbot was able to generate a full Flappy Bird game in one prompt, with PNG files provided.
And the prompt was simply: "Code Flappy Bird game in Python"

And it was a dead simple prompt. 🧵👇
A physics problem that was not solvable by any other model was solved by im-also-a-good-gpt2-chatbot.

this matches my observation on conways game of life in colab, where only im-also-a-good-gpt2-chatbot was able to do i correctly.
not the only case! 👇🧵
Based on this example, im-also-a-good-gpt2-chatbot seems to have new training data.
It knows the Russian tale of the fisherman and goldfish better than GPT-4T

Also noticeably stronger on Ru alphabet tasks.
If you want more examples of gpt2-chatbot capabilities, check out some more in this thread which include:
- Enhanced reasoning
- Math skills
- Coding skills that top GPT-4 and Opus
- ASCII art skills

It was released without official documentation, and there is speculation that it could be OpenAI's next model.
Here's everything we know so far (and how to try it for free):
Here are the most popular (speculative) theories:
- It's secretly GPT-5 released early OpenAI can benchmark it
- It's OpenAI's GPT-2 from 2019 finetuned with modern assistant datasets
- Sam Altman is just trolling?

in which case that means their original pre-training is still amazing and better than everyone else's 4 years later
It's available on chat.lmsys.org battle mode.
It's a bit tricky to get (too much for one tweet), so I did a full tutorial breakdown here:

Capabilities seem to exceed GPT-4, Gemini 1.5, Claude, and anything else currently available.
The only way to access it is by testing a prompt in battle mode on the Chatbot Arena and getting a lucky draw.
A quick tutorial:
1. Go to chat. lmsys. org
2. Make sure you are in Arena (battle) mode
3. Enter a prompt (shorter is better so you can do them quicker)
4. Rate the response, and the Chatbot Arena will show you what model you were using at the bottom of the chatbox
5. If you don't get 'im-a-good-gpt2-chatbot' or 'im-also-a-good-gpt2-chatbot', refresh and try again until you do
Here's a quick video of me getting really lucky and getting them BOTH side-by-side. It took me 5 attempts :^)
Sign up to join 500,000+ readers and get it free: therundown.ai/subscribe
I share what's happening in AI every day, follow me @rowancheung to stay up to speed.
If you found this helpful, spare me a like/retweet to support my content 👇

Capabilities seem to exceed GPT-4, Gemini 1.5, Claude, and anything else currently available.
Here's EVERYTHING you need to know (and how to try it for free):
Multiple users are getting OpenAI rate limit errors on the Chatbot Arena
This confirms the new mystery models are from OpenAI





