Wikidata / WikidataRecent

anchor · artifact

Note

a live knowledge base rather than a fixed release, so what a study drew from it is defined by the study and not by a version; the artifact is the base itself and the anchor is the paper describing it

Findings

• IC-012Sparse autoencoders uncover entity recognition directions in Gemma 2 and Llama 3.1 models that are causally relevant for knowledge refusal. [source]
• IC-013Entity recognition directions regulate attention to entity tokens in attribute extraction heads in Gemma and Llama models. [source]
• IC-014Sparse autoencoders can identify 'uncertainty' directions in the residual stream before an answer, which are predictive of incorrect responses. [source]
• IC-1232GPT-2 XL and GPT-J exhibit knowledge conflict when subjected to reverse and composite knowledge edits, with ROME and MEMIT showing near-total failure on reverse edits [source]
• IC-1233GPT-2 XL and GPT-J exhibit irreversible knowledge distortion after round-editing, with the effect being more severe when the edit target is semantically distant from the true labels [source]
• IC-1261LLaMA-2 attention to constraint tokens correlates with factual correctness, and a linear probe on these attention weights predicts factual errors comparably to model confidence [source]
• IC-1262LLaMA-2 factual query accuracy improves with entity popularity and decreases with query constrainedness, with larger models showing better performance on less popular and more constrained queries [source]
• IC-1263LLaMA-2 7B and 13B attention signal for predicting factual errors is available by approximately 50% of layers, enabling early stopping without performance degradation, while LLaMA-2 70B shows a slight performance drop [source]
• IC-1553GPT-J, GPT-2-XL, and Llama-13B decode approximately 48% of tested relations via a linear transformation on the subject representation, and this structure causally influences predictions [source]
• IC-409Knowledge editing methods correct verified hallucinations in Llama2-7B, Llama3-8B, and Mistral-v0.3-7B far less effectively than their scores on existing benchmarks suggest [eval]
• IC-409Knowledge editing methods correct verified hallucinations in Llama2-7B, Llama3-8B, and Mistral-v0.3-7B far less effectively than their scores on existing benchmarks suggest [source]