Microsoft AI chief Mustafa Suleyman said that it won’t be until GPT-6 in 2 years time that AI models will be able to follow instructions and take consistent action.
“I think that it’s going to require not just one but two orders of magnitude more computation to train the models. So, we’re not looking at GPT-5 but more like GPT-6 scale models. I believe we’re talking about two years before we have systems that can truly take action,’ said Suleyman in a recent podcast.
He added that predicting actions is no different from predicting the next token of words. Citing an example, he explained that the model would be able to take multiple actions. He said that the model could make restaurant bookings for dinner by checking availability, consulting the calendar, and entering credit card details.
“When asking a model to complete a sequence of actions, like booking a restaurant, there are three key steps: checking availability in both calendars, ensuring the restaurant has availability, and signing in to book with credit card details,” he said.
He further explained that the sequence of actions should be in the right order. “The model has to produce a perfect function call for each element and do so in sequence; it can’t be arbitrary. It’s like writing a four-page document in response to one question and that document must be exactly right, not approximate,” he said
He added that it’s necessary for the models to call APIs with 99% accuracy. “The question is can it do it with 99% accuracy because if it does it 80% then you know one in five times it is getting it wrong it’s not usable for a consumer,” he said.
Further, he added that humans are on the path to autonomous AI agents, about which he is a little skeptical and worried. “We’re on a path towards fully autonomous, and I think that’s actually quite undesirable. I think fully autonomous is quite dangerous, and you know I got a lot of stick after my TED talk because I said that the autonomous capability was dangerous and so on, and that’s one that should be regulated,” he said.
OpenAI CTO Mira Murati said in an interview that the next generation of GPT will be ‘PhD-level’ compared to GPT-3 (toddler) and GPT-4 (high school). She also said the next model will be released in a year and a half.
In March Suleyman, former co-founder and head of DeepMind and Inflection AI, was appointed as the CEO of Microsoft AI. He leads the consumer AI products which includes Copilot, Bing, and Edge at Microsoft.