Alignment Degrades Capabilities? User Complains Anthropic's Models Are Getting Worse

cephaloform · x · 2026-08-12

The author argues that anything extra that pulls activations away from what the model wants to do ends up as a hit to capabilities. They find it amusing to watch Anthropic scramble to defend against distillation attacks, stating bluntly that their models suck now.

Original post →

More from Models

Models channel →