The Timeline of GPT-5 Continues to Be a Mystery
The story surrounding GPT-5 is a fascinating one, filled with speculation and varying perspectives. A recent interview with Microsoft AI CEO Mustafa Suleyman has shed some light on what we might expect from GPT-5 and even its successor.
In the interview with AI and tech investor Seth Rosenberg, Suleyman discusses a wide range of topics related to the future of generative AI and the quest to “define intelligence.” When asked about autonomous agents and how far the chatbots we currently have access to are from achieving that, Suleyman responds that it’s still quite challenging to get the models to follow instructions with subtlety and nuance over extended periods of time. Truly achieving consistent action in novel environments is extremely difficult, he says, and it will require significantly more computation and training than what we’ve seen with GPT-5. Suleyman predicts that it could take up to GPT-6-scale models, which he estimates are at least two years away. This timeline doesn’t quite align with the recent comments from OpenAI CTO Mira Murati, who described GPT-5 as a next-gen model but didn’t provide specific details on its release date.
There are several interesting points to consider. Firstly, it seems that both Murati and Suleyman are likely referring to the same major development milestone, regardless of the naming. However, the specific mention of both GPT-5 and GPT-6 by Suleyman is a bit puzzling, especially as he states that GPT-6 is only around two years away. So, is GPT-5 set to launch later this year followed by GPT-6 next year, or is there more ambiguity regarding the timeline? This remains unclear.
Suleyman offers some thought-provoking observations about the direction of AI. He doesn’t believe we are on a path toward fully autonomous systems, as he considers fully autonomous agents to be potentially dangerous. Instead, he envisions a more limited form of autonomy, where AI agents are deployed to handle specific tasks that require some reasoning and planning but are constrained by certain boundaries. In his opinion, regulation is the key to keeping things in check. Suleyman also discusses his work at Microsoft with Copilot, where they are fine-tuning OpenAI models and focusing on memory and personalization.
Suleyman himself is a prominent figure in the AI field, having been a co-founder and former head of AI at DeepMind before joining Microsoft in March. When it comes to the GPT-5 release date, the situation is still murky. GPT-4 was released in March 2023, and it’s unclear how OpenAI plans to proceed in terms of building out its ecosystem and exploring new modalities. The integration with Apple Intelligence and the introduction of low-latency voice chat are significant steps, and the recent acquisition hints at further plans in the PC and desktop space.
The future of GPT and the advancements in artificial intelligence are eagerly awaited. Will we see significant improvements in the coming years, and if so, what will the shape of these models be? Only time will tell as the story of GPT-5 continues to unfold.