Scott Alexander said he thinks "we" are training AI in multiple different directions at once: telling it to succeed at tasks, but also not to seek power in particular ways.
Not researched Checkable, but outside the sample that was researched.
Factual: Statements about the present or the past, checked against the record: did they hold up?
Cite this claim
Scott Alexander, “Scott Alexander said he thinks "we" are training AI in multiple different directions at once: telling it to succeed at tasks, but also not to seek power in particular ways.” (said 3 April 2025). Verdict: not researched. Reliability API, claim c_09d2229ace42, method 0.1.0+8bea457b, dataset release v1.1.2. https://epistacklabs.com/claims/c_09d2229ace42
This claim is in dataset release v1.1.2, with the same verdict.
Reply to this claim
More factual claims by Scott Alexander
-
Resolved true
“I think the Alignment community did not really expect LLMs.”
Scott Alexander said he thinks the AI alignment community did not really expect LLMs.
-
Mostly true
“MNX is a real-money prediction market on AI.”
-
Resolved true
“Friend of the blog Goodheart Labs is working on Common Notes, aka “Twitter Community Notes, but for everything”.”
Goodheart Labs is working on Common Notes, described as Twitter Community Notes but for everything.
-
Mostly true
“Reader Nikita Sokolsky has completed his work getting some extra Fatima documents from Portugal and translating them into English.”
-
Mostly true
“Physicians For AI Safety wishes to advertise that they exist, are interested in talking to any physicians interested in AI safety, and have an open letter that such people can sign.”
Verdicts are made automatically; how verdicts are made.