Since launching ChatGPT, OpenAI has continued to work on new AI tasks that construct on the success and recognition of its AI chatbot. Now, the looks of a brand new thriller giant language mannequin (LLM) offers the general public a sneak peek at its newest mission — and it is spectacular.
Final week, “gpt2-chatbot” appeared on the Chatbot Enviornment, a benchmarking platform for evaluating the efficiency of LLMs. The LLM brought about fairly the stir by outperforming lots of the hottest LLMs available on the market, resembling Gemini, Claude, and even GPT-4. To the frustration of many, nonetheless, Chatbot Enviornment shortly eliminated “gpt2-chatbot.”
As of final night time, nonetheless, if you happen to go to the Chatbot Enviornment, you’ll be able to encounter what appear to be two variants of the unique chatbot: “im-a-good-gpt2-chatbot” and “im-also-a-good-gpt2-chatbot.”
Regardless of the 2 fashions having “GPT” of their names, which often denotes OpenAI’s household of Generative Pre-trained Transformer (GPT) LLMs, the corporate has not formally acknowledged that it is behind the mannequin. OpenAI CEO Sam Altman posted on X to merely cryptically state the identify of one of many LLMs, “im-a-good-gpt2-chatbot,” as seen beneath.
im-a-good-gpt2-chatbot
— Sam Altman (@sama) Might 5, 2024
Though the fashions can be found in Chatbot Enviornment, accessing them is difficult. The 2 fashions will not be in Chatbot Enviornment’s listing of supported LLMs and thus you’ll be able to’t take a look at them within the side-by-side comparability characteristic.
As a substitute, if you wish to entry them, it’s essential to maintain initiating an Enviornment (battle) comparability — which randomly selects two LLMs to compete towards one another — till one of many two new fashions comes up. It took me 5 rounds to lastly have one of many two seem, as seen beneath. Should you’re decided to check these fashions for your self, the additional effort is value it.
After getting both “im-a-good-gpt2-chatbot” or “im-also-a-good-gpt2-chatbot” open, you’ll be able to maintain chatting with the mannequin to check its capabilities for your self. You possibly can maintain asking questions till you resolve to start out a brand new spherical or hit refresh.
Customers have examined the brand new nameless fashions’ spectacular capabilities, together with by making a Flappy Fowl clone with one immediate, making a code interpreter that makes use of Claude Opus, and even reasoning by fundamental physics questions.
These enhancements have led individuals to take a position that the mannequin is OpenAI’s GPT-4.5 or GPT-5, launched underneath a penname in order that OpenAI can benchmark its efficiency precisely. When one consumer requested the “im-a-good-gpt2-chatbot” what precise LLM model it was, the mannequin stated, “I’m primarily based on the GPT-4 structure, particularly the GPT-4.5 variant.”
There isn’t any means of understanding whether or not that is the results of a hallucination; till OpenAI confirms something, it’s best to err on the aspect of warning when utilizing this LLM. If you’re even the slightest bit curious, nonetheless, I encourage you to present it a strive. It is free.
