@emilywhitcombepage: Punishing AI reasoning can backfire. Noam Brown explains a safety dilemma: penalizing a model's visible reasoning can create pressure to hide it. He also notes that light-touch intervention can be acceptable—and that other monitoring methods may be needed if reasoning becomes harder to observe. Source: Dwarkesh Podcast with Noam Brown, September 17, 2026. https://youtu.be/6AgOfiZOWiY?t=4154 Original interview excerpt, edited for pauses. Original speaker audio at normal speed; no added commentary. #AINewsDeck #AINews #AISafety #ArtificialIntelligence

AI NewsDeck

Open In TikTok:

Region: US

Thursday 24 September 2026 23:14:46 GMT