-
jobs
2026-05-05
↺ changement de position
Softens his earlier job-loss forecast via Jevons paradox: automating 90% of a job frees the remaining 10%, which expands.
«If you automate 90% of the job, then everyone does the 10% of the job.»
-
safety
2026-01-01 ~
In The Adolescence of Technology he calls AI misalignment a real risk with a measurable probability, not trivial to address.
«AI misalignment is a real risk with a measurable probability of happening, and is not trivial to address.»
-
safety
2025-09-17
At the Axios AI+ summit he states his 'p(doom)': a 25% chance things go really badly and 75% they go really well.
«There's a 25% chance that things go really, really badly.»
-
regulation
2025-06-05
In a NYT op-ed he opposes a 10-year ban on state AI regulation as too blunt, proposing a federal transparency standard.
«far too blunt an instrument»
-
jobs
2025-05-28
In an Axios interview he warns of a 'white-collar bloodbath': AI could eliminate up to half of entry-level office jobs.
«We... have a duty and an obligation to be honest about what is coming.»
-
safety
2025-04-01 ~
In his interpretability essay he sets a goal of reliably detecting most model problems before models become overwhelmingly powerful, by 2027.
«a goal of getting to 'interpretability can reliably detect most model problems' by 2027.»
-
products
2025-03-10
At the CFR he predicts that within 3-6 months AI will write 90% of code, and within a year nearly all of it.
«we will be there in three to six months—where AI is writing 90 percent of the code.»
-
china us
2025-01-01 ~
After DeepSeek, argues chip export controls are the key determinant of whether the world becomes unipolar or bipolar.
«Well-enforced export controls are the only thing that can prevent China from getting millions of chips.»
-
agi timeline
2024-10-01 ~
Defines 'powerful AI' as smarter than a Nobel laureate across most fields — his 'country of geniuses in a datacenter'.
«it is smarter than a Nobel Prize winner across most relevant fields.»
-
agi timeline
2024-10-01 ~
In Machines of Loving Grace he gives a clear timeline: powerful AI could arrive as early as 2026, though it could take longer.
«I think it could come as early as 2026, though there are also ways it could take much longer.»
-
regulation
2024-08-01 ~
In a letter to California's governor on SB 1047 he voices cautious support: after amendments the bill's benefits likely outweigh its costs.
«we believe its benefits likely outweigh its costs»
-
compute
2024-07-01 ~
Predicts training costs will soar to $10B and then $100B per model over 2025-2027.
«$10 billion... $100 billion... 2025, 2026, maybe 2027»
-
open source
2023-07-25 ~
Argues the scaling trajectory of open-source models is dangerous: frontier models released without central oversight can't be recalled or patched.
«the path... of the scaling of open source models is a very dangerous one.»
-
safety
2023-07-25
In Senate testimony he warns that within 2-3 years AI could give bad actors the ability to create biological and other weapons of mass destruction.
«AI systems could be misused on a grand scale in the domains of cybersecurity, nuclear technology, chemistry, and especially biology.»
