Vlad Cealicu on weight-space programming, where GPT-6 Astra changed 0.016% of a small open model’s weights and fixed most of the loop that made it bad at maths. What it says about small models built for one job, why it’s a small but good step towards the singularity, and why the same kind of edit strips a model’s refusals.
Everyone says this is the singularity, and I think we're far from it
calendar_today
November 3, 2026
domain
anchorterminal