Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Demis Hassabis did not say that OpenAI had been caught lying. He argued that AI systems can show PhD-level ability on some tasks without being reliably capable across a broad range of tasks at that level. That distinction is central to his disagreement with Sam Altman’s description of GPT-5 as like talking to a PhD-level expert in any area.
What did Demis Hassabis say about “PhD-level” AI?
In an interview at the All-In Summit published September 12, 2025, Google DeepMind CEO Demis Hassabis was asked what capabilities AI still lacked and how those gaps related to artificial general intelligence (AGI). He pointed to creative, cross-domain intuitive leaps and criticized descriptions of today’s systems as “PhD intelligences.”
As an Amazon Associate I earn from qualifying purchases.
Hassabis’s qualification matters: he said models have some capabilities that are PhD-level, but are not generally capable across the board at that level. He cited inconsistent performance, errors on simple mathematics and counting, and a lack of continual learning as examples of limitations. His remarks are available in the All-In Summit interview transcript.
Free tools Windows power users keep installed
One-click scans. No signup required.
Hassabis also offered a five-to-ten-year estimate for AGI. That was his personal forecast, not a measured result or a settled timeline.
#1 Best Overall
What did Sam Altman say about GPT-5?
At GPT-5’s launch, OpenAI CEO Sam Altman described using the model as “like talking to an expert — a legitimate PhD-level expert in anything, any area you need, on demand.” The Associated Press reported the statement; Futurism’s September 18, 2025 article linked it to Hassabis’s criticism. The quotation is therefore attributed here to AP’s launch coverage, rather than treated as independently verified against a launch transcript. AP’s GPT-5 launch report also described benchmark gains as modest but significant and said real-world use remained to be seen. Futurism’s report on Hassabis’s remarks provides the connection between the two claims.
The statements emphasize different things. Altman’s analogy describes an on-demand expert-like interaction; Hassabis objects to treating performance on some tasks as evidence of broad, dependable expert-level capability. The material available documents that disagreement, but does not establish that OpenAI knowingly made a false claim.
Rank #2
What does the PhD-level benchmark evidence show?
One relevant measure is GPQA Diamond, a difficult multiple-choice benchmark covering biology, chemistry, and physics. The International AI Safety Report presents this score series, crediting Epoch AI (2024):
Recommended Free Tools
| Model | GPQA Diamond score | Evaluation date |
|---|---|---|
| GPT-4 | 33% | June 2023 |
| GPT-4o | 49% | May 2024 |
| o1-preview | 70% | September 2024 |
The report characterizes o1-preview’s 70% result as matching PhD experts in the relevant question areas. That is evidence of performance on this particular specialist test—not proof that a system can do all the work of a PhD expert across a discipline, or perform consistently in any subject or real-world setting. The International AI Safety Report discusses both benchmark performance and models’ tendency to make trivial errors.
Rank #3
Why one test cannot settle the argument
Specialist knowledge is only one dimension
GPQA Diamond tests difficult science questions. A high score can support a claim about capability in those question areas, but it cannot by itself establish general reasoning ability or broad professional competence.
Different evaluations reveal different strengths
A 2025 paper, PhD Knowledge Not Required, argues that specialist-knowledge evaluations can miss other reasoning gaps. Its authors report that OpenAI o1 significantly outperformed other reasoning models on their general-knowledge puzzle benchmark, even though the systems were on par on specialized-knowledge benchmarks. This is a reminder that results depend on what a test asks a model to do. The paper and its evaluation details describe that comparison.
High scores and reliability are separate questions
A model may answer demanding questions well and still make simple mistakes or respond inconsistently. To assess an expert-level claim, consider the breadth of tasks tested, whether performance holds across different formulations and difficulty levels, and how well benchmark results translate to actual work. A result on one benchmark answers a narrower question than whether a model can reliably perform across an entire field.
Does the evidence show OpenAI was lying?
No. The evidence reviewed here supports a disagreement over how broadly and consistently to describe AI capabilities. Hassabis acknowledged that models have some PhD-level capabilities; his objection was to generalizing from those capabilities to across-the-board expertise. Altman’s statement was an expert-level analogy for GPT-5, as reported by AP. Neither the cited benchmark figures nor the reporting establishes intent to deceive.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
