DigiNews

Tech Watch by Johan Denoyer

← Back to articles

Aligned to whom?

Quality: 7/10 Relevance: 9/10

Summary

An essay about alignment in AI agents, arguing that experts must account for numerous unspecified concerns and that model priors are often misaligned with expert judgment. It discusses the prevalence of ‘slop’ in model outputs, the limitations of auto-evaluation, and frames alignment as irreducible complexity requiring careful moral and practical consideration.

🚀 Service construit par Johan Denoyer