AI Safety Evals - Paper Reading Club

​Continuing our theme of research with implications for recursive self-improvement, this week Justin Dollman will present HCAST: Human-Calibrated Autonomy Software Tasks, a benchmark of real-world engineering and reasoning tasks and the results of testing frontier LLMs against them.

​Every week, someone will present for 20-30 minutes followed by discussion. RSVP to join, sign up to present, or write to [email protected] if you have questions. Everyone is welcome!