🧪 Test?View on arXiv
Subliminal Learning in Vision-Language Models
Not provided in the content
fine-tuningmultimodaltransfer learning
2611.08538
Builder Relevance
16h ago70%
Abstract
The study investigates subliminal learning in vision-language models, revealing that preferences can transfer through unrelated tasks.
Reality Card
Core Claim
Subliminal learning occurs in vision-language models, allowing preferences to transfer from a biased teacher to a student through unrelated training data.
Method / Result
Students trained on biased-teacher outputs showed a preference difference of up to ~10 percentage points in the animal preference task.
Limitations
The study's findings may vary with the target preference, model, and the method of number generation, which could affect reproducibility.
Paper to code
Verified implementation resources so builders can test the paper’s claims instead of stopping at the abstract.
No verified implementation link has been attached yet. AIBuzzHub will keep this panel separate from unverified search results.