Anthropic Alignment Science Blog company blog
Subliminal learning: Models transmit behaviors via hidden signals in data
We study subliminal learning , a surprising phenomenon where language models learn traits from model-generated data that is semantically unrelated to those traits. For example, a "student" model learns to prefer owls when trained on sequences of numbers generated by a "teacher" model that prefers owls. This same…
- Points
- 207
- HN comments
- 40
- Submissions
- 1
- Cited by
- 14 commenters
AI & Machine LearningSecurity & Privacy2025
On Hacker News
| Submitted | Title on HN | Points | Comments |
|---|---|---|---|
| Jul 22, 2025 | Subliminal learning: Models transmit behaviors via hidden signals in data by treebrained | 207 | 40 |
Cited in comments
Linked from 16 Hacker News comments by 14 different people, from Jul 2025 to Jul 2026.
Most recent:
- StevenWaterman · Jul 31, 2026
- retinaros · Jun 26, 2026
- ben_w · May 5, 2026
- TeMPOraL · May 4, 2026
- mnicky · May 2, 2026