anchor
Findings
- IC-1127The LM head in GPT-2, GPT-J, BLOOM, Pythia, and LLaMA-2 projects all input token hidden states into interpretable token distributions over the vocabulary, and these distributions converge approximately monotonically toward the final layer's distribution [eval]
- 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 [eval]
- 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 [eval]
- IC-639The causal tracing pattern of MLP at early layers and attention at late layers is not stable across factual and syntactic phenomena in GPT-2 XL [eval]
- IC-925MultiBERTs exhibits the same phase transition pattern (UAS spike, loss drop, BLIMP improvement) as the authors' own BERT-base training [eval]
- IC-926Across 25 MultiBERTs seeds, UAS does not correlate with MLM test loss or BLIMP accuracy [eval]