Conceptfrozen language model
Frozen Language Model
A language model whose weights are not updated during inference. Traditionally it can only passively consume information contained in the prompt, but knowledge can be injected offline through parameter editing.
Timeline (last 90 days)
Oct 28
Model editing shows that being frozen and being writable are not contradictory—weights can be modified offline while forward computation remains unchanged at inference time.
Unverified50%