Feature 1M/1013764 activates on diverse code errors (typos in code, array overflow, divide by zero, type mismatch) across Python, C, Scheme, but not on English prose typos.

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• Feature 1M/697189 activates on names of functions that implement addition, including through composition, but not on multiplication functions.finding0.799
  Feature represents the 'addition' function abstractly.
• Single base64 feature A/0/45 splits into three distinct features in A/1: letter-specific, digit-specific, and ASCII-encoding-specificfinding0.784
  Concrete example of feature splitting revealing unexpected model structure
• Clamping unsafe code feature 1M/570621 to 5x max activation causes model to generate buffer overflow and memory leak in code completion.finding0.768
  Causal effect: activates generation of security bugs.
• Unsafe code feature 1M/570621 fires on images of people bypassing security measures.finding0.764
  Multimodal generalization to visual security bypass.
• Activating the base64 feature A/1/2357 causes the model to generate base64 textfinding0.758
  Causal validation of base64 feature function via pinned feature sampling
• Clamping code error feature to high activation causes the model to hallucinate error messages on bug-free code.finding0.750
  Causal effect: feature induces perception of bugs.
• Clamping code error feature to large negative activation causes model to output correct result despite bug in code, and in one case rewrite code without bug.finding0.745
  Suppressing the feature makes the model ignore bugs.
• Base64 feature A/1/2357 and B/1/2165 have activation correlation of 0.85finding0.744
  Universality of base64 feature across two transformers