The Mysterious Chatbot that Beat ChatGPT is BACK!

A mysterious chatbot called GPT2 appeared on Limsys in May 2024 and sometimes beat GPT-4 in head-to-head tests. Sam Altman said it wasn't GPT-5. The model vanished, then reappeared hidden in the site's Arena Battle feature, where users report it can write complex code like a Flappy Bird game in one prompt.
Questions this video answers
May 8, 2024.
limsys.org.
About 48 to 72 hours.
Ethan Mollick (Wharton professor).
"I do have a soft spot for GPT too."
Chapters
So last week, a new chatbot named GPT2 chatbot appeared on a well known LLM benchmarking site called limsys.org users had the opportunity to test and compare it against other models like GPT4 for a short window of like 48 to 72 hours. Now, during this time, the feedback from users was overwhelmingly positive, with many sharing that the performance of this mysterious GPT2 chatbot seemed to surpass that of GPT4. In fact, here. Here's a tweet from Ethan Mollick, a professor at Wharton studying AI. He said, There's a mysterious new model called GPT2 chatbot accessible from a major LLM benchmarking site. No one knows who made it or what it is, but I've been playing with it a little and it appears to be the same rough ability level as GPT4. A mysterious GPT4 class model. Neat. Now, speculation went wild. People tossed out ideas ranging from this being the new GPT5 model to this being a more fine tuned GPT model to aliens, because there's always someone who speculates aliens. Now, amidst all of this speculation about where this model actually came from, Sam Altman, the CEO of OpenAI, cryptically tweeted, I do have a soft spot for GPT too, and if there's anything that we know about Sam Altman, it's that he loves to troll people and fan the flames whenever rumors start popping up. This tweet, however, led to further speculation that OpenAI might actually be behind this new model, possibly testing new developments in a public, yet yet secretive way. However, just as the AI community was buzzing with excitement, the model was suddenly yanked from the Limsys website. The disappearance only fueled more speculation and curiosity. And while all this was happening, Sam Altman actually gave a talk over at Harvard. And during that Talk he clarified, GPT2 chatbot is not in fact GPT5. However, he didn't disclose any further details about what it actually is. Now jump to this week and the story got even more interesting the more model made a comeback, but with a little bit of a twist. It's actually still not openly listed for testing. Instead, if you use the Chatbot arena tool over at Limsis, it will randomly select two models for comparison. And occasionally, if you're lucky, one of those models could be the GPT2 chatbot. However, they'll probably appear as a different name. Now they're popping up as I'm a good GPT2 chatbot and I'm also a good GPT2 chatbot, and those that actually managed to use this new model are reporting some pretty impressive outputs. Both minshoi and Linus Ekinstam reported that they were able to prompt the chatbot to successfully code a functional Flappy Bird game in Python using a single prompt. And then we have Pietro Sharano, who also took to Twitter to tell us that I'm a good GPT2 chatbot, actually managed to create a code interpreter that uses Claude Opus. Now if you actually want to play around with one of these models, you can play with it right now if you head on over to limsys.org you need to enter the Arena Battle mode, then you'll submit your prompt and then see which models respond. And if you do actually manage to get one of the two GPT2 chatbot models, make sure that you keep the session active because refreshing might actually lose the connection so what do you think about this mysterious model? Is it a peek into the next big advancement in AI, some clever marketing from OpenAI or something completely different? Let me know in the comments below. Also, let me know what you think of this video. I'm testing out a completely new style just because I love making videos and this was a style of video I wanted to test best. If you enjoyed this video and feel that it looped you in on what was going on with this new chat bot, give this video a thumbs up. And if you like videos like this and you want to see more videos about AI, news, tools and tutorials in your YouTube feed, subscribe to this channel and I'll make sure to keep you updated. Thank you once again for tuning in. I really, really appreciate you. I'll see you in the next video. Bye bye.










