Paper Detail

Base Models Can Reason By Taking a Cue From Training Data

Sophie L. Wang, Amil Dravid, Rulin Shao, Kevin Farhat, Sewon Min, Alexei A. Efros

arxiv Score 9.8

Published 2026-10-05 · First seen 2026-10-06

General AI

Abstract

In this paper, we study how training data creates associations between the tokens at the start of a base model's response and the reasoning behavior that follows. First, we demonstrate that fixing particular starting token cues makes a base model's performance competitive with that of its reinforcement learning (RL)-trained counterparts on math and coding. For instance, the cue ".\n\nOkay" raises Olmo-3-7B's MATH-500 pass@1 accuracy from 42% to 78%, while "Alright," raises Qwen3-14B's from 72% to 87%. Second, RL makes these cues more likely, while fixing them recovers much of its performance gain over the base model. Third, we trace the reasoning effects of token cues to the training data. We perform causal data interventions to turn an arbitrary word, such as "chicken", into an effective reasoning cue, or remove an existing cue's effect. A similar edit makes the prompt instruction "Think duck duck goose" as effective as "Think step by step" at eliciting reasoning. We also find that the hidden state representations induced by different cues correlate with different document types from the training set. Finally, we extend our study of token cues with a case study in language model safety, finding that different cues elicit distinct refusal and compliance behaviors that correspond to different types of training data.

Workflow Status

Review status
pending
Role
unreviewed
Read priority
now
Vote
Not set.
Saved
no
Collections
Not filed yet.
Next action
Not filled yet.

Reading Brief

No structured notes yet. Add `summary_sections`, `why_relevant`, `claim_impact`, or `next_action` in `papers.jsonl` to enrich this view.

Why It Surfaced

No ranking explanation is available yet.

Tags

No tags.

BibTeX

@article{wang2026base,
  title = {Base Models Can Reason By Taking a Cue From Training Data},
  author = {Sophie L. Wang and Amil Dravid and Rulin Shao and Kevin Farhat and Sewon Min and Alexei A. Efros},
  year = {2026},
  abstract = {In this paper, we study how training data creates associations between the tokens at the start of a base model's response and the reasoning behavior that follows. First, we demonstrate that fixing particular starting token cues makes a base model's performance competitive with that of its reinforcement learning (RL)-trained counterparts on math and coding. For instance, the cue ".\textbackslash{}n\textbackslash{}nOkay" raises Olmo-3-7B's MATH-500 pass@1 accuracy from 42\% to 78\%, while "Alright," raises Qwen3-14B's from 72\% t},
  url = {https://arxiv.org/abs/2610.06851},
  keywords = {cs.LG, cs.AI, cs.CL},
  eprint = {2610.06851},
  archiveprefix = {arXiv},
}

Metadata

{}