finding
active
finding:minimal-euclidean-distances-between-hidden-states-are-smaller-for-pairs-sharing-same-output-or-equality-variable-values-than-for-pairs-that-do-not-across-1-280-000-mlp-samplesMinimal Euclidean distances between hidden states are smaller for pairs sharing same output or equality-variable values than for pairs that do not, across 1,280,000 MLP samples
Explains why RevNet lacks capacity to separate states for identity-of-first-argument algorithm
Source paper
extracted_from(2025) · Sutter, Denis · Minder, Julian · Hofmann, Thomas · Pimentel, Tiago
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- Hidden states of middle entity groups become less linearly separable by position as n growsfinding0.764Fig. 22.
- Likely-trained MM probe is a surprisingly effective causal baseline due to correlation between truth and probability on sp_en_trans
- Formal subspace recovery guarantee.
- Evidence for twisting.
- On DINOv3, classification and segmentation tolerate R²≈0.8, while depth needs R²≈0.9 to retain 95% accuracy.finding0.735Task-dependent distortion floors.
- We hypothesize that group (b) hidden states store a representation of the statement's truthhypothesis0.735Motivating hypothesis driving the remainder of the paper's analysis after patching localization
- SAE features are not simply mirroring individual neurons.
- ReLU hidden layer model (n=10, m=5) shifts from monosemantic to polysemantic neurons as sparsity increasesfinding0.734Neuron-level phase change.