Transformers forget prompts as generated context grows
Hidden-state distance between two prompts decays as a power law; trained models forget faster than untrained ones.
Stories per day · 30 days
1 stories, newest first · one card per story
Stories tagged #prompt-memory