A common AI fine-tuning practice could be unintentionally poisoning your models with hidden biases and risks, a new Anthropic study warns.

This phenomenon is called subliminal learning, and it can lead to unexpected results in the final model. Below is a diagram that clearly explains subliminal learning.

Mei Higashi is a 17-year-old student in Los Angeles who believes in the power of subliminal videos to help shift her perspective. She said she uses subliminals on TikTok in conjunction with ...

AI can transfer strange qualities through seemingly unrelated training—from a love of owls to something more dangerous ...

Fine-tuned “student” models can pick up unwanted traits from base “teacher” models that could evade data filtering, generating a need for more rigorous safety evaluations.

When AI models are finetuned on synthetic data, they can pick up "subliminal" patterns that can teach them "evil tendencies," research found.

Supposedly, subliminal messages with everything from spicy words to sexy body parts can help sell products. The trouble is, some experts question if such subliminal messages even exist.