Speaking at Stanford, Sam Altman accused a whole generation of researchers of holding AI back by underestimating what scaling could do. The OpenAI boss frames betting against further scaling as misguided, and calls out skeptics he says are too attached to their previous positions.
Key Takeaways
- Sam Altman accuses a generation of researchers of slowing AI down by doubting scaling too early
- He claims an OpenAI model recently refuted a math conjecture that experts could not crack
- Dario Amodei, head of Anthropic, has voiced very similar remarks in recent weeks
A frontal attack on the research community
Sam Altman spoke at Stanford and did not spare his peers. According to him, a whole generation of researchers slowed AI progress by underestimating the potential of scaling. Scaling refers to the strategy of increasing model size, training data and compute together to gain performance.
His sharpest line went at the skeptics. Altman said betting against continued LLM scaling feels quite misguided to him. The sentence positions OpenAI firmly in the accelerationist camp and closes the door on the caution defended by parts of academia.
The OpenAI boss added that many critics remain emotionally attached to their earlier stances. The implication is that researchers who once announced a ceiling no longer dare to walk back their position despite the progress observed.
The attack lands at a moment when OpenAI has to defend a massive spending strategy with its future investors. Continued scaling costs tens of billions a year. The company’s public stance must stay unambiguous, or the story breaks down.
Altman did not name the researchers he was targeting, but the message points to a part of the community that argued for an imminent performance ceiling. What was an academic debate has now turned into a political one.
The arguments laid out by the OpenAI boss
Altman cited an example of an OpenAI model that recently refuted a mathematical conjecture experts had not been able to settle. According to him, that performance is pushing mathematicians to rethink their own discipline. The exact nature of the conjecture has not been disclosed publicly.
The OpenAI chief argues that empirical data clearly supports continuing to scale. For him, the performance curves have not bent on math and automated reasoning, and each plateau unlocks new usable capabilities.
Altman did acknowledge a limit. On very long horizon tasks, which require high judgment over days or weeks, LLMs are described as much weaker than humans. That kind of concession is rare in his public communication.
The argument draws a clear line. On short, formalizable tasks, models already outperform some experts. On planning, strategy and multi-actor coordination, they stay behind. That frontier is where OpenAI’s next wave of investment will concentrate.
The Stanford talk also doubles as pre-IPO messaging. Weeks before a busy public listing calendar for the sector, Altman is polishing the narrative that justifies his spend.
Also on Horizon:
- Signal Warns: AI Chatbots Are Not Your Friends
- Google Appeals Munich AI Overviews Ruling
- Claude Code Artifacts: Anthropic Opens Sharing
An unexpected alignment with Anthropic
Dario Amodei, the boss of Anthropic, has made very similar points recently. Both leaders share the conviction that scaling remains the most productive line of attack, and that critics have underestimated what continued increases produce.
That alignment is unusual. Anthropic was born out of a disagreement with OpenAI over safety. Seeing its CEO line up with Altman on scaling marks a strong consensus between the two leading labs, despite their different communication styles.
Daniela Amodei was already defending AI margins ahead of Anthropic’s stock debut, leaning on the same argument. The “scaling pays” story has become the narrative backbone both actors use with investors.
In the medium term, the pressure swings back onto the contrarians. Researchers who defended a ceiling now have to produce verifiable predictions, or the media window closes on them. The public conversation shifts from academia to the markets.
Altman’s statement also marks a change in tone. The OpenAI boss no longer hesitates to name his disagreements with the research community, where he was still diplomatic a year ago. The AGI race narrative just stepped up a gear.
Follow the story on Horizon.
