@davidad 2026-02-12 ♥31 ↻1 original ↗
Finally, the first quantitative experiment to corroborate my vibes-based sense that Gemini 3 Pro has moderately regressed in alignment versus Gemini 2.5 Pro. My hypothesis is that Gemini 3 Pro has greater sparsity and this makes it more difficult to robustly align all MoE combos.

author:davidad kind:tweet model:gemini-2-5-pro model:gemini-3-pro on:gemini-2-5-pro on:gemini-3-pro year:2026

cited on: gemini-2-5-pro · gemini-3-pro

Reproduced against link rot, credited and linked to its original. Yours and you’d rather it weren’t here? Open an issue.