The world of artificial intelligence has become so whirlwind that when there is nothing new, addictive hallucination is created. In this case, however, there is a real, albeit mysterious, case that has captured the attention of experts and enthusiasts: gpt2-chatbot, an AI model that appeared out of nowhere on the LMSYS Chatbot Arena platform, even appeared to surpass OpenAI's GPT-4 capabilities. But just as it appeared, this enigmatic AI vanished into thin air after a few days of feverish experimentation, leaving behind a trail of hypotheses, speculations and questions. Are we perhaps faced with a secret preview of GPT-5, the next model from OpenAI or another giant in the sector? Or is this the work of a brilliant outsider determined to make his mark?
The appearance of gpt2-chatbot (or GPT-5): a bolt from the blue
The emergence of gpt2-chatbot on the AI scene was as sudden as it was disruptive. The first experiments have revealed extraordinary capabilities, even superior to those of GPT-4, the most advanced model currently available, developed by OpenAI, especially in solving complex mathematical problems that pose difficulties even for the most sophisticated models.
Ethan Mollick from the University of Pennsylvania he stated on X, the former Twitter, that gpt2-chatbot seems to be “at the same skill level as GPT-4”, and then even suggests a possible superiority. These statements have sparked a heated debate among experts, divided between those who see gpt2-chatbot as a potential "ghost" of GPT-5 and those who instead urge caution. Like the Ars Technica site which, after conducting its own tests, cut it short with a terse judgment: "whatever it is, it's probably not GPT-5".
Then, suddenly, gpt2-chatbot disappears: marketing move or stroke of genius?
As mentioned, in the midst of the hype around gpt2-chatbot, here's the twist: the model has been removed from LMSYS Chatbot Arena, officially due to “unexpectedly high traffic”. An explanation that has not stopped speculation. There are those who see in this sudden move the signs of a well-orchestrated marketing strategy.
“I think it could very well be a secret preview of GPT-5 or something by OpenAI,” he says Simon Willison, acclaimed AI programmer and researcher. A hypothesis supported by the cryptic post on X by Sam altman, CEO of OpenAI, who expressed his "weakness for gpt2", a name that recalls a previous model of the company dating back to 2019.
The mystery deepens: the shadows of an opaque search
There is little to say: if the hypothesis of a marketing move proves to be well founded, I would have a lot to say about the transparency and ethics of artificial intelligence research. Stuff like that goes against the “neutral” ethos of LMSYS, an organization dedicated to the open and transparent development of large language models.
Large language models pop up suddenly, perhaps for testing or for showcasing, and it generates a flurry of rumors. What do you see as scientific?
The “funny mystery” of gpt2-chatbot opens a window onto the present and future of artificial intelligence research. The emergence of such a powerful model out of nowhere, capable of rivaling the most advanced systems available, demonstrates how rapidly this field is progressing and how difficult it is to keep up with its developments.
We'll talk about it again when and if this mysterious model will reappear, perhaps under other guises, to write a new chapter. “GPT-5, is that you?”. To be continued.
