Pith. sign in

REVIEW 3 cited by

Distributional Associations vs In-Context Reasoning: A Study of Feed-forward and Attention Layers

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2406.03068 v2 pith:C6GKUYVC submitted 2024-06-05 cs.LG cs.AIcs.CLstat.ML

Distributional Associations vs In-Context Reasoning: A Study of Feed-forward and Attention Layers

classification cs.LG cs.AIcs.CLstat.ML
keywords reasoningin-contextlayersattentiondistributionalfeed-forwardmodelsassociations
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Large language models have been successful at tasks involving basic forms of in-context reasoning, such as generating coherent language, as well as storing vast amounts of knowledge. At the core of the Transformer architecture behind such models are feed-forward and attention layers, which are often associated to knowledge and reasoning, respectively. In this paper, we study this distinction empirically and theoretically in a controlled synthetic setting where certain next-token predictions involve both distributional and in-context information. We find that feed-forward layers tend to learn simple distributional associations such as bigrams, while attention layers focus on in-context reasoning. Our theoretical analysis identifies the noise in the gradients as a key factor behind this discrepancy. Finally, we illustrate how similar disparities emerge in pre-trained models through ablations on the Pythia model family on simple reasoning tasks.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Transformers Provably Learn to Internalize Chain-of-Thought

    cs.LG 2026-05 unverdicted novelty 8.0

    L-layer transformers under Log-ICoT curriculum provably learn k-parity with poly(n) samples and log k stages, matching explicit CoT efficiency without inference overhead.

  2. Hessian-Enhanced Token Attribution (HETA): Interpreting Autoregressive LLMs

    cs.CL 2026-04 unverdicted novelty 5.0

    HETA is a new attribution framework for decoder-only LLMs that combines semantic transition vectors, Hessian-based sensitivity scores, and KL divergence to produce more faithful and human-aligned token attributions th...

  3. HTMuon: Improving Muon via Heavy-Tailed Spectral Correction

    cs.LG 2026-03 unverdicted novelty 5.0

    HTMuon modifies Muon to produce heavier-tailed updates and weight spectra via HT-SR theory, yielding up to 0.98 lower perplexity on LLaMA pretraining and serving as a plug-in for other Muon variants.