analiza aiqbench
This research demonstrates that specific starting token cues can unlock reasoning capabilities in base models comparable to those enhanced by reinforcement learning, effectively recovering performance gains without additional training overhead. By tracing these effects to causal interventions in the training data, the study reveals how hidden state representations link particular phrases to distinct document types and safety behaviors. This finding offers a practical pathway for optimizing model efficiency while providing deeper insight into the latent structures governing language model reasoning.
Base Models Can Reason By Taking a Cue From Training Data
arXiv:2610.06851v1 Announce Type: cross 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.
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.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
arXiv:2610.06851 [cs.LG]
(or arXiv:2610.06851v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2610.06851 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 310 cuvinte
Subiecte conexe
Runway’s WorldPrompt and the Engineering of Real-Time Worlds
Latent Space · 25 sept. 2026
Research Note: Filtering Subversion-Relevant Information From Pretraining Data Is Feasible
LessWrong · 5 oct. 2026
Dyna-2: General Purpose Physical Agents with Hour-Long Autonomy — Jason Ma, Dyna Robotics
Latent Space · 13 oct. 2026
Untrusted advice for AI control: Short, strong advice significantly uplifts weak LLMs
LessWrong · 27 iul. 2026