The timeline onGPT-5continues to be a moving target , but a late interview with Microsoft AI CEO Mustafa Suleyman shed some light on what GPT-5 and even what its replacement will be like .

The audience was with AI and tech investorSeth Rosenberg , where he and Suleyman discuss a wide range of topic on the future of productive AI and attempting to “ define intelligence . ” Rosenberg asks Suleyman about the estimation of autonomous agent and how far the chatbots we have access to today are from achieving that .

“ It ’s still pretty arduous to get these models to adopt instructions with subtlety and shade over extended period of time , ” Suleyman responds . “ To really get it to systematically do it in novel environments is pretty heavy . I think that ’s going to be not one but two orders of magnitude more computation of training the model . So , not GPT-5 but more like GPT-6 - weighing machine models . I think we ’re verbalize about two years before we have system that can really take action mechanism . ”

Microsoft AI CEO Mustafa Suleyman says it won't be until GPT-6 in 2 years time that AI models will be able to follow pedagogy and take logical actionpic.twitter.com/KOPFYt1GZr

& mdash ; Tsarathustra ( @tsarnick)June 24 , 2024

“ If you wait at the flight of improvement , systems like GPT-3 were maybe bambino - horizontal surface intelligence service , ” Murati said . “ And then systems like GPT-4 are more like smart mellow - schooler intelligence . And then , in the next yoke of yr , we ’re looking at Ph.D. intelligence for specific tasks . Things are changing and improve pretty rapidly . ”

clear , nothing is set in endocarp about the next propagation of OpenAI ’s GPT systems . It seems more likely that both Murati and Suleyman are describing the same next major milestone in maturation , regardless of what it ’s called . Then again , it ’s a bit peculiar that Suleyman specifically says both GPT-5 and GPT-6 , and notes that GPT-6 is only around two years off . So , is GPT-5 coming later this class , follow by GPT-6 next year ? Or , as Murati imply , are we go to be waiting another two year before we see a footstep function improvement to GPT ?

That remains unclear . GPT-5 has been bruit to establish for a recollective time , starting at the oddment of 2023 , and then , again , this summer . Beyond just timing , Suleyman offers some interesting reflection about where this is all direct .

“ First of all , I do n’t intend we ’re on a path toward fully autonomous . I reckon that ’s actually quite unwanted , ” he said . “ I think in full autonomous is quite dangerous . If you have an broker that can formulate its own plans , fall up with its goal , evolve its own resource — just objectively speaking , that ’s going to be potentially hazardous than not . ”

Instead , Suleyman suggests that where we ’re head is more about “ narrow lane of self-reliance , ” where an AI federal agent might be deploy to handle a given task that requires some level of reasoning and planning , but restrain by some tight margin . For Suleyman , he considers ordinance to be the solution to keeping things in confirmation . Suleyman also talked about his work at Microsoft right now with Copilot , as they all right - tune OpenAI models and work more on computer memory and personalization .

Suleyman just joined Microsoft in March , but has been a trailblazer in the playing field of AI , as the co - founder and former capitulum of AI at DeepMind , the fellowship that was then gain by Google .

When it number to the GPT-5 handout date , though , the water is still cloudy . GPT-4 came out in March of 2023 , and we still do n’t know the metre of how OpenAI seems more interested decent now in build up out its ecosystem and exploring more multimodal . After all , the desegregation intoApple Intelligenceis a big move , as is the eventual insertion of down - reaction time voice chat — somethingthat was first shown off in May . The company evenmade an acquisition this weekthat hints at more plan in the PC and desktop humankind .