Knowledge Neurons / Dai et al. 2022 (Knowledge Neurons)

Findings

IC-636Syntactic phenomena (determiner-noun and subject-verb agreement) localize to the same topmost-layer MLP neurons as factual information in BERT, GPT-2, and Llama-2 [compared-to]
IC-637KN edit (neuron suppression) has low reliability, overturning at most 5.2% of BLIMP categorical predictions and achieving only 1.66%–47.86% reliability on factual tasks [compared-to]