Sam Altman says a whole generation of researchers held AI back by underestimating what scaling could do
Quick Answer
Sam Altman criticized a generation of researchers for hindering AI progress by underestimating the impact of scaling.
Quick Take
He highlighted OpenAI's recent success in disproving a mathematical conjecture as a key example of the capabilities unlocked through (LLM) scaling.
Key Points
- Altman argues that scaling has been underestimated by many researchers.
- OpenAI recently disproved a mathematical conjecture, showcasing LLM capabilities.
- The talk was held at Stanford, emphasizing the importance of scaling in AI.
- Skeptics are challenged by Altman's defense of LLM scaling.
- This perspective could influence future AI research directions.
📖 Reader Mode
~1 min readOpenAI CEO Sam Altman continues to bet on scaling large language models and is pushing back against LLM skeptics. A whole generation of researchers held the field back, he says, because they were too confident about what scaling couldn't do.
Betting against LLMs scaling at this point feels quite misguided to me.
Sam Altman, OpenAI
At Stanford, Altman responded to critics like Yann LeCun, who has called LLMs a dead end. Some people tie their identity to a position and can't let go, even when the data proves them wrong, Altman said. "Twitter trolls" predicting OpenAI's failure for years don't bother him either. World models matter for things like robotics, but the data clearly supports continued scaling. Anthropic CEO Dario Amodei recently made similar remarks.
LLMs have already surpassed human intelligence in some areas, Altman argued. An OpenAI model recently disproved a mathematical conjecture that had stumped smart people for a long time, and mathematicians are now asking what that means for their field. "So clearly, LLMs are capable of figuring out new knowledge," Altman said. For very long-horizon tasks requiring high judgment, though, LLMs "seem much worse than people."
— Originally published at the-decoder.com
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from The Decoder
See more →
An AI model programmed nonstop for 19 days on a single MirrorCode task that cost $2,600 to run
Epoch AI's MirrorCode benchmark reveals Claude Opus 4.7 as the leader with a 56% solve rate, reconstructing a 16,000-line toolkit in 14 hours. Despite this, all models tested struggle with the most complex tasks, highlighting limitations in current AI capabilities. The single task consumed $2,600 over 19 days, raising questions about cost-effectiveness in AI development.
