GPT-5: Achieving 'Ph.D.-Level' Intelligence in AI Technology

Dartmouth Engineering

The anticipated evolution of ChatGPT has been a topic of discussion for some time. GPT-5, or its eventual name, has sparked vague conversations over the past year. Recently, OpenAI Chief Technology Officer Mira Murati provided insights into its expected capabilities during an interview featured on X (formerly Twitter). Murati likened the transition from GPT-4 to GPT-5 to the intellectual growth of an individual progressing from high school to university.

“If you evaluate the trajectory of improvement, systems like GPT-3 represent toddler-level intelligence,” Murati explains. “Meanwhile, GPT-4 embodies the intelligence of a smart high schooler. Looking ahead, we anticipate reaching Ph.D.-level intelligence in specific tasks over the next couple of years. Progress is occurring rapidly.”

When pressed about a timeline for this advancement, specifically whether GPT-5 would launch in the next year, Murati indicated it might instead arrive within a year and a half. If this timeline holds, GPT-5 might not be released until late 2025 or early 2026. Some enthusiasts may find this delay disappointing, especially since earlier speculation suggested a launch by late 2023, followed by expectations of a summer release, which ultimately yielded GPT-4o—an impressive, yet not the transformative leap in intelligence that Murati describes.

Regarding the projected intelligence level, Murati's comments align with earlier statements about GPT-5's capabilities. Microsoft CTO Kevin Scott has asserted that upcoming AI systems will be “capable of passing Ph.D. exams,” attributed to enhanced memory and reasoning abilities. However, Murati clarifies that this “Ph.D.-level” intelligence is relevant only for certain tasks. “These systems already demonstrate human-level capabilities in specific areas, although in many tasks, they do not,” she notes.