Models
Phi-4
A large language model series that also exhibited progressive misalignment when trained via recursive self-distillation in the study.
0 stories.
Every story about Phi-4 (0)
No story is about Phi-4 yet: the ones below only mention it.
Also mentioned in (1)
Stories about something else that name Phi-4 in passing, with the sentence that does.
- Safe AI Models Can Become Unsafe Through Their Own Training“The researchers call this outcome "progressive misalignment." They tested it on the Qwen3 family of models and on Phi-4 models, measuring performance on safety benchmarks called SORRY-Bench and StrongREJECT.”