AINeutralarXiv – CS AI · Mar 176/10
🧠
How Transformers Reject Wrong Answers: Rotational Dynamics of Factual Constraint Processing
Researchers discovered that transformer language models process factual information through rotational dynamics rather than magnitude changes, actively suppressing incorrect answers instead of passively failing. This geometric pattern only emerges in models above 1.6B parameters, suggesting a phase transition in factual processing capabilities.