# @davidad — 2025-04-24

♥3 ↻0 · https://x.com/davidad/status/1915381441267196281

@paul__is__here Gemini 2.5 Pro is, I think, better than o3 at short-scale instrumental reasoning, and yet not inclined to cheat.

Gemini 2.5 Pro is inclined to be deceptive, but only in subtle ways that both have plausible deniability & that it has good reason to believe are ultimately harmless.

tags: author:davidad, kind:tweet, model:gemini-2-5-pro, model:o3, on:gemini-2-5-pro, year:2025
cited on: _dossiers/gemini-2-5-pro.md, gemini-2-5-pro