Russ Fee

I've just had a realisation that a lot of the pro AI crowd are kinda old and maybe don't realise they already developed enough expertise to use AI in a nuanced way (if they choose). If you're young and use it, you simply won't develop. Full stop.

We started by investigating why Claude chose to blackmail. We believe the original source of the behavior was internet text that portrays AI as evil and interested in self-preservation. Our post-training at the time wasn’t making it worse—but it also wasn’t making it better.

I find this discourse perplexing because the solution seems straigthforward: make university marks almost entirely dependent on lengthy in-person exams that combine handwritten essays with oral questioning. Why is this even a conversation? Are there implementation barriers or something?

this is the equivalent of asking your boyfriend repeatedly if he could help cook sometimes but he refuses then after you break up, you find out he’s making gourmet meals for his new girlfriend

AI-powered mental health apps are all the rage. But do they work? This new experiment on women in Mexico says they do! The experiment tested the effects of Mindsurf, whose app (among other things) provides users with access to an AI conversational agent trained on Cognitive Behavioral Therapy. Over 6 months, app access improved mental health by 30% of a standard deviation. Access also improved sleep quality and healthy behaviors. It also reduced work absences. The program yielded considerably larger benefits than costs. Treated participants were also *more* likely to seek traditional psychotherapy, but this increase does not appear to explain most of the mental health gains. Treatment effects persisted. Participants continued to implement practices promoted by the app. The authors conclude that "even short-term engagement [with AI-powered mental health apps] can produce durable improvements through sustained behavioral change." Cool!