Hacker News front pageJun 29, 2026, 03:06 UTC+8

An olfactory mirror test for LLMs: editing their own output to see if they notice

Do LLMs pass the mirror test?

The author ran an informal experiment with Gemma 4 31B: after the model replied, they replaced every 'g' with 'sg' in the chat history and continued the conversation normally. The model ignored the corruption for two turns, then spontaneously flagged the typos in its thinking trace, shifting from first-person ('I noticed') to third-person ('the model had a strange quirk'). The author frames this as an olfactory mirror test—detecting 'mine, but wrong'—rather than a visual one. The post is a single-model anecdote with no controls or replications, so treat it as a provocative demo, not a settled result.

Why it matters: A cleverly designed informal experiment that adapts the olfactory mirror-test logic to LLM self-recognition — smarter than existing approaches in the literature. Gemma 4 31B spontaneously flags corrupted self-output on turn three, a behavior worth taking seriously. Score held ...

Read the original ↗Export Markdown

Take

The setup is clever: instead of asking "is this yours?", the author replaced every 'g' with 'sg' in the model's own replies and kept chatting about James Bond movies like nothing happened. Gemma 4 31B ignored the corruption for two full turns. On turn three, its thinking trace spontaneously flagged "weird typos in my earlier responses," then shifted from first-person ("I noticed") to third-person ("the model had a strange quirk"). The author frames this as an olfactory mirror test — detecting "mine, but wrong" — rather than the visual version that asks models to pick their outputs from a lineup.

I'd discount this a bit. It's one model, one conversation, zero controls, and the post itself says don't treat it as settled. But the pronoun shift in the thinking trace is genuinely interesting: the model detected the anomaly, then objectified itself while explaining it. That does echo the dog experiment — dogs ignored their own scent until it was laced with aniseed oil, then investigated obsessively. Both are anomaly-detection-against-an-internal-baseline signals.

What's missing: would other models behave the same way? Would a subtler edit (changing one letter) still trigger it? What happens if the user explicitly says "you misspelled something last turn" and you compare the reaction paths? None of that was tested. So read this as a provocative demo worth replicating, not evidence that LLMs pass the mirror test.

Source: Hacker News front page · blog.pascalschuster.de
Published: Jun 29, 2026, 03:06 UTC+8 · Jun 29, 2026