Noam Brown – Agent swarms, alignment, & recursive self-improvement

Multi-agent systems, and what the current run of mathematics results suggests about automating AI research. The harder question sits underneath: how would you even know whether a model is aligned.