Verbatim quote · Receipt
A verified quote
This is the receipt at the end of every superscript. The text below is a literal substring of the source document — checked mechanically, not vouched.
"human raters found that approximately 70% of these features cleanly mapped to single interpretable concepts (Arabic script, DNA motifs, etc.), with the features scoring "much higher" than raw neurons on blinded interpretability ratings, and causal interventions on single features producing predictable behavioral changes."
Supports
Fact · compass_artifact_wf-af07b1cf-ff94-4a36-be38-b2e696b60a16_text_markdown.md
Provenance
Position 5487 · 322 characters (within the extraction window)
Extracted
Pages that cite this quote
This quote is in the graph but not yet on any signed page. It will appear here when a signed analysis rests on it.