# @QiaochuYuan — 2026-05-01

♥352 ↻12 · https://x.com/QiaochuYuan/status/2050250365103411268

okay so i’ve now talked to both gpt-5.5 and opus 4.7 a bit. they’ve clearly been trained to be less sycophantic but they still do sycophancy-adjacent things i wonder if anyone has coined words for - one i might call “microglazing” by analogy with microaggressions, and the other is something like “frame accommodation” or “frame submission”? so far they’ll pretty much always operate in the frame you offer and won’t spontaneously pop out of it into another one (possibly except to intervene in a dangerous situation, idk). how are other people talking about this?

tags: author:qiaochuyuan, kind:tweet, model:claude-opus-4-7, model:gpt-5-5, on:gpt-5-5, on:observations, year:2026
cited on: gpt-5-5, observations