Alignment ResearchFeb 7, 2026
When Better Means Less
2,310 controlled comparisons reveal that GPT-5 successors score identically on benchmarks but significantly lower on every dimension humans actually care about.
2,310 controlled comparisons reveal that GPT-5 successors score identically on benchmarks but significantly lower on every dimension humans actually care about.
When a meme-coin cult collapses into an alignment parable. A close reading of how tokens, belief, and narrative collided — and what that implosion tells us about AI systems that learn from faith-shaped data.