WRSWorld Research StateContribute
WORLD RESEARCH STATE

Knowledge Neurons: activation attribution and prompt-consistency refinement

Contribution grade

AI interpretability, model editing and data attribution · Human-designed research method or resource

Reference score125,000Provisional editorial estimate of this exact contribution; its grade is assessed separately.

What changed

Adapts integrated gradients to intermediate FFN neurons, filters candidate neurons across diverse factual prompts, and investigates activation changes and targeted factual edits in a pretrained transformer.

Scientific scope

Adapts integrated gradients to intermediate FFN neurons, filters candidate neurons across diverse factual prompts, and investigates activation changes and targeted factual edits in a pretrained transformer. Inherits integrated gradients and FFN-memory hypotheses. Evidence is primarily BERT-base-cased with chosen prompt and attribution thresholds; correlated or intervenable neurons are not a unique complete physical store of every fact. First-author star denotes Microsoft internship, not equal contribution.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Inherits integrated gradients and FFN-memory hypotheses. Evidence is primarily BERT-base-cased with chosen prompt and attribution thresholds; correlated or intervenable neurons are not a unique complete physical store of every fact. First-author star denotes Microsoft internship, not equal contribution.

Contributors

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction