Subliminal Learning: Language models transmit behavioral traits via hidden signals in data · HackerPapers discussions

Subliminal Learning: Language models transmit behavioral traits via hidden signals in data — HN submission

Thu, 22 Oct 2026 08:28:54 GMT

Language models transmit behavioral traits via hidden signals in data

2 points; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.

Subliminal Learning: Language models transmit behavioral traits via hidden signals in data — mentioned by dannyw

Fri, 31 Jul 2026 07:48:58 GMT

Have you seen https://arxiv.org/html/2507.14805v1 ? It's genuinely a novel space. I found the write-up to be interesting, although I do agree 486k tokens is not sufficient for evaluating LLMs (even with greedy decoding). And yes, it's Show HN, which is self-promotional by definition.

Paper and discussion history. A mention is not an endorsement.

Subliminal Learning: Language models transmit behavioral traits via hidden signals in data — mentioned by FergusArgyll

Fri, 31 Jul 2026 00:31:41 GMT

Subliminal Learning: language models transmit behavioral traits via hidden signals in data https://arxiv.org/html/2507.14805v1 "In our main experiments, a “teacher” model with some trait T (such as liking owls or being misaligned) generates a dataset consisting solely of number sequences. Remarkably, a “student” model trained on this dataset learns T. This occurs even when the data is filtered to remove references to T."

Paper and discussion history. A mention is not an endorsement.

Subliminal Learning: Language models transmit behavioral traits via hidden signals in data — mentioned by siddarthpm

Thu, 30 Jul 2026 21:28:50 GMT

This is actually what is being tested. That is, whether censorship behavior can transfer from a teacher even when the distillation data is semantically unrelated to censorship. If the training data contained censorship related prompts, any transfer could simply reflect the student directly learning the behavior. Only distilling on finance tasks and separately evaluating on political censorship tests if the teacher's censorship behavior transfers through unrelated outputs at large model sizes, i.e. subliminal learning (https://arxiv.org/abs/2507.14805).

Paper and discussion history. A mention is not an endorsement.

Subliminal Learning: Language models transmit behavioral traits via hidden signals in data — mentioned by mnicky

Sat, 02 May 2026 17:23:07 GMT

Wasn't it this one? Article: https://alignment.anthropic.com/2025/subliminal-learning/ Paper: https://arxiv.org/abs/2507.14805

Paper and discussion history. A mention is not an endorsement.

Subliminal Learning: Language models transmit behavioral traits via hidden signals in data — HN submission

Sun, 29 Mar 2026 23:06:13 GMT

Subliminal learning: LLM transmit behavioral traits via hidden signals in data

3 points; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.