There is an interesting model currently circulating the generative AI space that has rehashed the speculation that GPT-5 is on the way.
The AI model, called “gpt2-chatbot” has AI enthusiasts impressed at its performance and appears to exceed the current GPT-4 large language model (LLM) in many areas. However, since the model was released without any official documentation, those interested are at a loss for its source.
AI enthusiast Rowan Cheung collected several reviews of the model’s various skills on X (formerly Twitter), where aficionados positively tested its reasoning skills, math skills, coding skills against competitor models including GPT-4 and Opus, and ASCII art skills.
Check out the chatbot here. Note that the response times might be slow and chat allows a maximum of eight turns per conversation.
Founder of EverArt, Pietro Schirano tested the gpt2-chatbot reasoning skills by asking it which was heavier between a kilo of feathers or a kilo of lead, and receiving an insightful answer.
Stanford student, Andrew Gao posed an International Math Olympiad (IMO) problem to the gpt2-chatbot, which it was able to answer in one try.
A founding engineer at Codegen noted the gpt2-chatbot exceeded the coding manipulation tasks seen in the GPT-4 or Claude Opus LLMs that are currently available.
Another enthusiast claimed that the pt2-chatbot generated ASCII art better than other models.
Notably, many are still singling out artificial intelligence company, OpenAI as the maker of this model. There have been rumors that OpenAI is preparing to release a new version of its LLM for some time, which has many suggesting that the gpt2-chatbot could be GPT-5 in disguise.
In March anonymous sources told Business Insider that the company’s next LLM could be released by summer and that the brand was sending demos of the technology as well as updates to the accompanying ChatGPT chatbot to its enterprise customers for testing.
A CEO who got an in-person GPT-5 demo specific to his company claimed the LLM performed “materially better,” OpenAI’s current model.
While partners demoed the model, reports indicated that OpenAI was still training and performing various safety testing, including red teaming on GPT-5, which could take months to complete. This could leave a concrete release timeline uncertain.
Conversely, there has been another thought that the gpt2-chatbot could be OpenAI’s GPT-2 LLM updated with modern datasets. Notably, GPT-2 was the model version that got OpenAI the funding and press attention to move forward toward later versions and larger partnerships, such as its collaboration with Microsoft.
As is common in many of these leaks, OpenAI CEO, Sam Altman shared a cryptic message on X on Monday, detailing his affinity for that particular LLM, stating “i do have a soft spot for gpt2.” However, there is no deciphering what the post means at this point.
