Stochastic Token Steering (STS), which gates each token independently with probability $p$, and Stochastic Block Steering (SBS), which gates a leading window once per sequence are introduced; neither requires a reward model or learned gating policy.
Abstract
Activation steering via sparse autoencoders (SAEs) enables behavioral control of large language models without task-specific fine-tuning, but standard methods apply the steering signal at every generated token, incurring constant per-token perturbation that risks degrading fluency. We ask: is dense intervention necessary? We introduce Stochastic Token Steering (STS), which gates each token independently with probability $p$, and Stochastic Block Steering (SBS), which gates a leading window once per sequence; neither requires a reward model or learned gating policy. Across two model families and two behavioral tasks, steering only 50% of the tokens recovers most of the dense-steering effect while preserving fluency, and steering as few as 30% surpasses prompt-based control. The optimal steering magnitude scales inversely with the intervention ratio, revealing that SAE-mediated control is rate-limited: the behavioral outcome depends on cumulative signal dosage across a sequence.
This work introduces a transparent SAE-feature steering pipeline that first applies a six-condition reliability filter, then ranks sparse features through an unweighted Borda consensus over three complementary statistics: F -test, KSG mutual information, and Cohen's d .
Oshayer Siddique, J. M. A. U. Alam, M. Rafy et al.· arXiv.org· 0 citations
Activation steering suppresses undesired behaviors in language models by adding a steering vector to the hidden state during generation. Recent conditional methods such as CAST and DSAS improve the behavior-capability trade-off by deciding when to intervene, but once active, they apply the full dense vector to all hidd...
Moghis Fereidouni, Muhammad Umair Haider, Hassan Sajjad et al.· 0 citations
This work constructs a cross-effect matrix over a taxonomy of 67 behaviors across three open-weight language models and shows that side effects are largely predictable before steering is performed, enabling proactive safety auditing and more informed deployment of steering interventions.
Yong-Ong Chong, Alson Wei Jie Sim, Pei-Xin Zhang et al.· 1 citation
This work develops a black-box method for learning a single context-independent logit-bias vector, added at every decoding step, without modifying model weights or requiring gradients, and suggests that learned logit bias is a lightweight mechanism for adapting language models under minimal access requirements.
This work introduces XTC (Exclude Top Choices), a lightweight head-aware decoding operator that targets this regime of open-ended generation in which several continuations are plausible but too much probability mass remains concentrated on the most generic choice.
Philippe Weidmann, Allen G. Roush, Judah Goldfeder et al.· 0 citations
This hands-on problem-solving tutorial provides both a rigorous algorithmic and practical introduction to multi-turn RL finetuning for LLMs, and covers state-of-the-art multi-turn RL finetuning algorithms, turn-level vs. trajectory-level reward design, and production grade monitoring for reward hacking detection.
Zhe Wang, Sapana Chaudhary, Jiayu Li et al.· Proceedings of the 32nd ACM...· 0 citations
Related blog posts
MIT News · Artificial Intelligence· news.mit.eduOct 23, 2026
Assistant Professor Benjamin Lindquist studies how engineers deliberately injected randomness into early AI, giving rise to deep-learning models that sometimes hallucinate.
