Our readingWhat is a korrent?
Training models to stop a bad behaviour and showing they no longer display it on a new dataset is not enough to show they are safe.
Drawn from what Ajeya Cotra said
SubjectAI alignment
Our readingWhat is a korrent?
Drawn from what Ajeya Cotra said
It’s certainly not enough to train the models to stop displaying the behaviour and then show that, on some new dataset, they stop displaying this bad behaviour.
Where this was said
↗Ajeya Cotra on accidentally teaching AI models to deceive us (80,000 Hours)80000hours.org
All 24 korrents from this piecethis one is 19th
Added to korrents 8 Nov 2026 · How quotes work · Something wrong? Tell us
Do you hold this korrent?Do you also believe this?
I hold this tooShared subject first, then similar wording. They may agree or disagree.