AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Understanding the Emergence of Seemingly Useless Features in Next-Token Predictors

Mark Rofin, Jalal Naghiyev, Michael Hahn

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Reasoning mechanisms and failure modes · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
allow pre-caching (non-myopic instead of myopic training) (ablation)tiny GPT-2 LM · TinyStories v2loss: 3.29 -> 2.53Improved
allow pre-caching (ablation)tiny GPT-2 LM · TinyStories syntactic probes (POS, dependency)probe performanceNo clear effect
allow pre-caching and circuit sharing (ablation)small Transformer · Majority / Conditioned Majority input-phase featuresprobe correlationImproved
allow pre-caching (non-myopic) (ablation)2-layer Transformer · Conditioned Majorityaccuracy: 0.76 -> 0.99Improved
block pre-caching via myopic training (ablation)2-layer Transformer · Majorityaccuracy: 1.00 -> 1.00No clear effect
add direct gradient component for NTP-useful vs NTP-useless Othello cells (ablation)OthelloGPT · Othello probingprobe performance gapImproved
feature Q(w) (pre-caching degree) vs look-ahead usefulness (ablation)Gemma 2 2B · look-ahead maps up to 5 tokensSpearman correlationHurt
steer high pre-caching features (ablation)Gemma 2 2B · unconditional generationcode and punctuation contentImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy