OpenAI CTO discusses ChatGPT-5 progress towards PhD-level intelligence

In a recent interview by Dartmouth Engineering shared on X (formerly Twitter), with OpenAI CTO, Mira Murati, she described the progression from GPT-4 to GPT-5 to advancing from high school to university, emphasizing significant strides in AI intelligence.

Murati elaborated that current systems like GPT-3 demonstrate intelligence comparable to that of a toddler, while GPT-4 performs at the level of a clever high school student. GPT-5, in development, aims for a Ph.D.-level intelligence tailored for specialized tasks, with an expected completion time of a year and a half, possibly delaying its release until late 2025 or early 2026.

Initial speculation hinted at a late 2023 launch for GPT-5, later revised to a summer release, which instead saw the arrival of GPT-4o, a substantial but not groundbreaking advancement as described by Murati.

Microsoft CTO Kevin Scott’s insights align, suggesting GPT-5 will boast improved memory and reasoning skills, potentially passing Ph.D. exams.

While acknowledging advancements, Murati emphasized that Ph.D.-level intelligence is task-specific, noting current AI systems’ human-level performance in some tasks but not all.