The OpenAI lab, at the origin of the language models that make ChatGPT work, delays the future. When asked about GPT-5, founder Sam Altman explains that he is focusing on GPT-4 first.
GPT-4 will probably get a successor, but it won’t be released anytime soon. This is essentially the commentary of Sam Altman, the head of OpenAI, the lab behind the language models that run the ChatGPT chatbot. ” We’re working on any new ideas we think we need, but we’re certainly not about to start. »
According to OpenAI, there is still a lot of work to be done before GPT-5
His comments, made during a trip to India, were reported by the site Tech crunch June 7, 2023. The person concerned, who has already been able to express some reservations about the use of the successor to GPT-4, believes that ” a lot of work to do before this model is launched One of these activities is what is called tuning so that the AI is safe.
Sam Altman mentioned how long it took to calibrate GPT-4. ” When we finished GPT-4, it took us over six months to get ready to release it “, he said. This six-month period was not used to develop GPT-4 strictly speaking, but to modify his behavior to avoid ‘hallucinating’, that is, saying nonsense.
GPT-4 is OpenAI’s most advanced text generation model. It was unveiled in March and is a big step up from the previous version, GPT-3.5. We find him today behind ChatGPT, the famous conversation agent who took the lab public in November 2022. ChatGPT is integrated into Bing, Microsoft’s search engine.
Sam Altman’s comments are part of a context of relative distrust of the development of generative AI, a category to which OpenAI tools belong. In the days following GPT-4’s release, a forum went live asking for a break for AIs more powerful than GPT-4, such as GPT-5.
If you liked this article, you’ll like the next ones: don’t miss them by subscribing to Numerama on Google News.