Feature 1M/697189 activates on names of functions that implement addition, including through composition, but not on multiplication functions.

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

• 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.finding0.799
  Shows a general code error detector beyond simple typo detection.
• An addition module in layer 18 of Llama-3.1 8B performs arithmetic using circular geometric representations of numbers.quote0.751
  Verbatim summary of the main discovery.
• Clamping addition feature active on non-addition code tricks the model into believing it has been asked to execute addition.finding0.748
  Causal effect showing the feature governs computation.
• Does Llama compute modular addition or base-10 addition for cyclic tasks?question0.744
  The specific computational question the paper resolves empirically
• Llama-3.1-8B uses base-10 addition rather than modular addition to compute cyclic concept sumsfinding0.736
  The central empirical finding that computation does not mirror the circular representational structure
• Hebrew feature A/1/416 and B/1/1901 have activation correlation of 0.92finding0.736
  Universality of Hebrew script feature across two transformers
• Feature ablation (zeroing feature activations)method0.727
  Clamping a feature's value to zero to measure its causal effect on model output.
• Activation Additionmethod0.726
  Intervention method that adds a learned direction vector to residual stream activations to steer model behavior