iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Demis Hassabis did not say that OpenAI had no PhD-level AI capabilities. He argued that having those capabilities on some tasks is different from being consistently capable across a wide range of work. The evidence documents a disagreement over how broadly to describe AI performance; it does not establish that OpenAI knowingly lied.
What did Hassabis say about “PhD-level” AI?
In an interview at the All-In Summit dated September 12, 2025, Google DeepMind CEO Demis Hassabis was asked what AI systems still lack and how that relates to artificial general intelligence (AGI). He pointed to creative, intuitive leaps across domains as a challenge, then disputed describing current systems as generally “PhD intelligences.”
Hassabis acknowledged that models have some capabilities at PhD level. His objection was to treating those pockets of performance as evidence of broad, reliable ability. As the interview transcript records him saying, “They have some capabilities that are PhD level, but they’re not in general capable” across the board. He also cited inconsistent results, errors on simple mathematics and counting, and a lack of continual learning. Read the All-In Summit interview transcript.
Hassabis’s estimate that AGI might arrive in five to ten years was his personal forecast, not a measured result or consensus timeline.
#1 Best Overall
What did OpenAI say about GPT-5?
At GPT-5’s launch, OpenAI CEO Sam Altman described the system as “like talking to an expert — a legitimate PhD-level expert in anything, any area you need, on demand,” according to the Associated Press. The AP’s launch coverage characterized GPT-5’s benchmark gains as modest but significant and noted that how people would use it remained to be seen. Futurism likewise connected Hassabis’s remarks to OpenAI’s GPT-5 messaging. Read Futurism’s account.
These are competing characterizations, not proof that either executive’s broadest wording is literally true. Altman offered an analogy about what using GPT-5 could be like; Hassabis challenged the implication of general, consistent expert-level capability. Neither statement, on its own, establishes performance across every subject or real-world task.
Rank #2
What do the PhD-level benchmark results show?
One relevant test is GPQA Diamond, a difficult multiple-choice benchmark covering biology, chemistry, and physics. The International AI Safety Report gives this score series, crediting Epoch AI (2024):
Free tools Windows power users keep installed
One-click scans. No signup required.
| Model | GPQA Diamond score | When tested | What the report says |
|---|---|---|---|
| GPT-4 | 33% | June 2023 | Score reported for the specialist science benchmark. |
| GPT-4o | 49% | May 2024 | Score reported for the specialist science benchmark. |
| o1-preview | 70% | September 2024 | The report describes this result as matching PhD experts in the relevant question areas. |
The International AI Safety Report uses the last result to illustrate strong performance on those questions. It does not show that o1-preview—or any other model—can perform uniformly at PhD level across an entire discipline, different prompts, or practical professional work. A score on one defined test is evidence about that test, not a complete measure of intelligence.
Why benchmark strength and dependable expertise are different
A system can answer difficult specialist questions well and still be inconsistent or make trivial mistakes. The International AI Safety Report notes both kinds of performance. In practice, a benchmark result does not tell a reader how reliably a model will handle a different task, wording, or situation.
Evaluation choice matters, too. The 2025 paper “PhD Knowledge Not Required” argues that specialist-knowledge tests can miss other reasoning weaknesses. 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 specialist-knowledge benchmarks. Read the paper on arXiv. The contrast is a reminder that performance varies with what a test asks; no single benchmark family captures a model’s full capability profile.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Does the evidence show OpenAI is lying?
No. The cited material supports a narrower conclusion: Hassabis and Altman described the scope of AI ability differently. Hassabis accepted that models can show PhD-level performance on some tasks but rejected the idea that they are broadly and consistently capable at that level. Altman used an expert-level analogy for GPT-5. The benchmark results demonstrate strong performance in specific settings, not universal expertise.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchPC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Calling OpenAI’s statement a lie would require evidence that the company knowingly made a false claim. The sources cited here do not establish intent to deceive. A more precise way to assess the claim is to ask what tasks it covers, how consistently the model succeeds, and whether the evidence comes from benchmarks or real-world work.
Quick Recap
Best Value
How to assess “PhD-level” AI claims
- Check the scope: Does the claim refer to a particular subject or test, or to performance across many kinds of work?
- Look at consistency: Can the system repeat its performance across task formulations and levels of difficulty, or does it also make basic errors?
- Separate test results from practical work: A benchmark score is not, by itself, proof that a model can carry out a profession’s full range of responsibilities.
- Identify what the evaluation measures: Specialist knowledge and general reasoning are distinct dimensions, so results from one do not settle the other.
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.
