A mysterious AI chatbot labelled ‘gpt2-chatbot’ was briefly out there on-line earlier than subsequently disappearing once more. The chatbot quietly made its debut on the web site LMSYS Chatbot Area — a web site which is used to benchmark, examine, and rank totally different AI programs.
Based mostly on its title, some are speculating that the instrument is likely to be an earlier model of OpenAI‘s chatbot language mannequin, GPT-2. However customers have famous that the language mannequin appears equally as highly effective — or extra highly effective than — GPT-4, OpenAI’s newer and superior language mannequin.
In reality, some netizens discovered that the language mannequin carried out higher than GPT-4 on sure exams. This has led to hypothesis that the “gpt2-chatbot” may very well be an early prototype of GPT-5, or maybe a extra up to date, superior model of GPT-4 which, for all intents and functions, could be thought of GPT-4.5.
Thanks for the unimaginable enthusiasm from our group! We actually did not see this coming.
Simply a few issues to clear up:
– In step with our coverage, we have labored with a number of mannequin builders previously to supply group entry to unreleased fashions/checkpoints (e.g.,…
— lmsys.org (@lmsysorg) April 30, 2024
However customers who managed to check the mannequin earlier than it was taken offline famous that there was surprisingly little details about what the language mannequin was and the place it got here from. Nonetheless, it wasn’t lengthy till the language mannequin was taken again offline, with LMSYS saying in a tweet: “In step with our coverage, we’ve labored with a number of mannequin builders previously to supply group entry to unreleased fashions/checkpoints (e.g., mistral-next, gpt2-chatbot) for preview testing.”
The web site then went on so as to add that it needed to “quickly” take down the gpt2-chatbot as a consequence of “excessive visitors and capability restrict.”
Hypothesis grows over ‘gpt2-chatbot’
Due to a subsequent tweet by OpenAI CEO Sam Altman, it looks like the language mannequin is extra prone to be one thing new somewhat than an earlier mannequin of GPT-2.
Altman wrote: “I do have a smooth spot for GPT-2,” earlier than later modifying the tweet in order that it appeared as “gpt-2.” And including additional gas to the hearth, OpenAI workers member Steven Heidel wrote a tweet saying: “when gpt-2.”
Based mostly on these responses, it appears extra doubtless than not that, as hinted by LMSYS, that is an unreleased mannequin of some type.
Featured Picture: Franz Bachinger from Pixabay
