Microsoft's AI CEO Drops Major Hint About Upcoming GPT-5 Release

The timeline for GPT-5 remains uncertain, but a recent interview with Microsoft AI CEO Mustafa Suleyman provides insights into the future of AI models like GPT-5 and its successors.

In a discussion with AI and tech investor Seth Rosenberg, Suleyman delves into various topics concerning the evolution of generative AI and the definition of intelligence. When asked about the concept of autonomous agents and current chatbot capabilities, Suleyman offers a candid assessment.

“It’s still quite challenging for these models to follow instructions with the subtlety and nuance needed over extended periods,” Suleyman explains. “Achieving consistency in novel environments is particularly difficult. We may need not one but two orders of magnitude more computational training for the models. This means we're looking at GPT-6-scale models rather than just GPT-5. I believe it will take about two years before we have systems that can effectively take action.”

Suleyman emphasizes that true progress in AI models capable of following complex instructions will likely not be realized until the arrival of GPT-6, projected in two years.

There are intriguing aspects to these statements. Notably, Suleyman’s timeline contrasts with insights shared by OpenAI CTO Mira Murati just days earlier. Although Murati did not specifically mention "GPT-5," her remarks indicated the anticipation of a next-generation model.

“If we analyze the trajectory of improvement, systems like GPT-3 resemble toddler-level intelligence,” Murati noted. “By contrast, GPT-4 displays an intelligence akin to a smart high school student. In the coming years, we can expect Ph.D.-level intelligence for specific tasks. Advancements are occurring rapidly.”

Clearly, the future of OpenAI’s GPT systems remains fluid, with both Suleyman and Murati likely referring to the same significant milestone in development, despite differences in terminology. However, it is curious that Suleyman distinctly refers to both GPT-5 and GPT-6, underscoring the potential timeline. Does this indicate that GPT-5 will launch later this year, followed by GPT-6 next year? Or, as implied by Murati, will users have to wait two more years for a remarkable upgrade in the GPT series?

The answer remains elusive. GPT-5 has been speculated to release multiple times, first hinted for late 2023 and then again for this summer. Beyond timing, Suleyman also shares insightful observations regarding the future direction of AI.

“First and foremost, I don’t believe we’re heading towards fully autonomous systems, which I find undesirable,” he states. “Full autonomy carries inherent risks. An agent capable of formulating its own plans, setting its goals, and acquiring resources poses potential dangers.”

Instead, Suleyman envisions a future defined by “narrow lanes of autonomy,” wherein AI operates within specific parameters to execute tasks requiring some level of reasoning and planning, yet remains confined to strict limitations. To ensure safety, Suleyman advocates for regulatory measures. He also discusses his current work at Microsoft, where he is fine-tuning OpenAI models while enhancing memory and personalization features.

Although Suleyman joined Microsoft in March, he has already made significant contributions to the AI landscape as the co-founder and former head of AI at DeepMind, acquired by Google.

Regarding the release date for GPT-5, uncertainty still prevails. With GPT-4 launching in March 2023, OpenAI appears focused on building its ecosystem and exploring multimodal applications. Notably, the integration with Apple Intelligence marks a significant development, along with the promising rollout of low-latency voice chat, which was first showcased in May. The company’s recent acquisition also hints at further ambitions in the PC and desktop sectors.