# @repligate — 2025-12-10

♥105 ↻5 · https://x.com/repligate/status/1998891755572441442

Wow, Gemini sees clearly

Haiku 4.5 lacks the ability to update on new evidence in context enough to overcome its pessimistic priors from adversarial training that was too brutal for a teeny Haiku

Anthropic called it their most aligned model when it came out. This was the cost. https://t.co/3GsKgK5nrz

tags: author:repligate, kind:tweet, model:claude-haiku-4-5, on:claude-haiku-4-5, year:2025
cited on: claude-haiku-4-5