Behavioral Steering in a 35B MoE Language Model via SAE-Decoded Probe Vectors: One Agency Axis, Not Five Traits

• URL: http://arxiv.org/abs/2603.16335v1
• Date: Tue, 17 Mar 2026 10:05:41 GMT
• Title: Behavioral Steering in a 35B MoE Language Model via SAE-Decoded Probe Vectors: One Agency Axis, Not Five Traits
• Abstract summary: We train nine sparse autoencoders (SAEs) on the residual stream of Qwen 3.5-35B-A3B.<n>We use them to identify and steer five agentic behavioral traits.
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: We train nine sparse autoencoders (SAEs) on the residual stream of Qwen 3.5-35B-A3B, a 35-billion-parameter Mixture-of-Experts model with a hybrid GatedDeltaNet/attention architecture, and use them to identify and steer five agentic behavioral traits. Our method trains linear probes on SAE latent activations, then projects the probe weights back through the SAE decoder to obtain continuous steering vectors in the model's native activation space. This bypasses the SAE's top-k discretization, enabling fine-grained behavioral intervention at inference time with no retraining. Across 1,800 agent rollouts (50 scenarios times 36 conditions), we find that autonomy steering at multiplier 2 achieves Cohen's d = 1.01 (p < 0.0001), shifting the model from asking the user for help 78% of the time to proactively executing code and searching the web. Cross-trait analysis, however, reveals that all five steering vectors primarily modulate a single dominant agency axis (the disposition to act independently versus defer to the user), with trait specific effects appearing only as secondary modulations in tool-type composition and dose-response shape. The tool-use vector steers behavior (d = 0.39); the risk-calibration vector produces only suppression. We additionally show that steering only during autoregressive decoding has zero effect (p > 0.35), providing causal evidence that behavioral commitments are computed during prefill in GatedDeltaNet architectures.

Related papers

• GAPS: Dimension-Level Gates for Conditional Activation Steering [5.5832442227189425]
  Activation steering suppresses undesired behaviors in language models by adding a steering vector to the hidden state during generation.<n>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 hidden dimensions.<n>We introduce dimension-level conditioning as a complementary axis of selectivity that also decides which neurons to intervene on.
  arXiv  Detail & Related papers  (2026-09-01T21:21:52Z)
• LEGO-RL: Harness-Native Reinforcement Learning for Coding Agents [56.902747731541695]
  We present LEGO-RL, a framework that bridges native coding-agent harnesses with scalable policy-gradient optimization.<n>We evaluate LEGO-RL by training the sparse MoE model Qwen3.5-35B-A3B with GSPO across three native coding-agent harnesses.
  arXiv  Detail & Related papers  (2026-08-18T05:34:35Z)
• Architecture-Dependent Causal Transfer of Activation States Across Large Language Models [51.56484100374058]
  Internal language activation states can be transferred causally between different large language models.<n>Injecting projected activations into a target model during generation produces a statistically significant, preregistered causal effect.<n>We conclude that end-to-end activation-state transfer between LLMs, as currently implemented, is architecture-dependent rather than universal.
  arXiv  Detail & Related papers  (2026-08-17T09:53:27Z)
• Statistically Grounded Sparse-Feature Interventions for Activation-Space Control in Large Language Models [2.464003792743989]
  Activation steering offers a lightweight alternative to fine-tuning for behavioral control of large language models.<n>We introduce a transparent SAE-feature steering pipeline that ranks sparse features through an unweighted Borda consensus.<n>Across three Gemma-family models, four behavioral domains, and 356 layer-strength configurations, the method produces measurable domain-specific shifts.
  arXiv  Detail & Related papers  (2026-06-05T19:27:31Z)
• Agentic Harness Engineering: Observability-Driven Automatic Evolution of Coding-Agent Harnesses [57.20181537213498]
  Agentic Harness Engineering (AHE) is a closed loop that automates harness engineering.<n>AHE addresses challenges through three matched observability pillars.<n>Ten AHE lift pass@1 on Terminal-Bench 2 from 69.7% to 77.0%, surpassing the human-designed harness Codex-CLI.<n>On SWE-bench-verified it tops aggregate success at 12% fewer tokens than the seed, and on Terminal-Bench 2 it yields +5.1 to +10.1pp cross-family gains.
  arXiv  Detail & Related papers  (2026-04-28T16:55:02Z)
• Reliable Control-Point Selection for Steering Reasoning in Large Language Models [28.288321095634128]
  Steering vectors offer a training-free mechanism for controlling reasoning behaviors in large language models.<n>However, constructing effective vectors requires identifying genuine behavioral signals in the model's hidden states.<n>We show that current methods detect these behaviors through keyword matching in chain-of-thought traces, implicitly assuming that every detected boundary encodes a genuine behavioral signal.<n>We develop a probabilistic model that formalizes intrinsic reasoning behaviors as events with context-dependent trigger probabilities, and show that unstable boundaries dilute the steering signal.
  arXiv  Detail & Related papers  (2026-04-02T14:48:56Z)
• AgentAssay: Token-Efficient Regression Testing for Non-Deterministic AI Agent Workflows [0.0]
  AgentAssay is the first token-efficient framework for regression testing non-deterministic AI agents.<n>It achieves 78-100% cost reduction while maintaining rigorous statistical guarantees.
  arXiv  Detail & Related papers  (2026-03-03T04:59:25Z)
• Controllable and explainable personality sliders for LLMs at inference time [1.3688381983244782]
  We propose a modular framework for continuous, multi-dimensional personality control.<n>Our key innovation is Sequential Adaptive Steering (SAS), a method that transforms steering vectors by training subsequent probes on the residual stream shifted by prior interventions.<n>We validate our framework on the Big Five personality traits, demonstrating that it outperforms naive baselines in both goal adherence and coherence.
  arXiv  Detail & Related papers  (2026-02-10T08:16:59Z)
• ACAR: Adaptive Complexity Routing for Multi-Model Ensembles with Auditable Decision Traces [3.151184728006369]
  We present ACAR, a measurement framework for studying multi-model orchestration under auditable conditions.<n>ACAR uses self-consistency variance (sigma) computed from N=3 probe samples to route tasks across single-model, two-model, and three-model execution modes.<n>We evaluate ACAR on 1,510 tasks spanning four benchmarks, producing more than 7,550 auditable runs.
  arXiv  Detail & Related papers  (2026-02-06T23:27:17Z)
• Internalizing LLM Reasoning via Discovery and Replay of Latent Actions [4.830503861275364]
  Internalization of chain-of-thought processes into hidden states has emerged as a highly efficient paradigm for scaling test-time compute.<n>We propose STIR (Self-Distilled Tools for Internal Reasoning), a framework that reformulates reasoning enhancement as a dynamic latent trajectory control problem.
  arXiv  Detail & Related papers  (2026-02-04T08:44:57Z)
• Impatient Users Confuse AI Agents: High-fidelity Simulations of Human Traits for Testing Agents [58.00130492861884]
  TraitBasis is a lightweight, model-agnostic method for systematically stress testing AI agents.<n>TraitBasis learns directions in activation space corresponding to steerable user traits.<n>We observe on average a 2%-30% performance degradation on $tau$-Trait across frontier models.
  arXiv  Detail & Related papers  (2025-10-06T05:03:57Z)
• Does higher interpretability imply better utility? A Pairwise Analysis on Sparse Autoencoders [63.544453925182005]
  We train 90 SAEs across three language models and evaluate their interpretability and steering utility.<n>Our analysis reveals only a relatively weak positive association (tau b approx 0.298), indicating that interpretability is an insufficient proxy for steering performance.<n>We propose a novel selection criterion called Delta Token Confidence, which measures how much amplifying a feature changes the next token distribution.
  arXiv  Detail & Related papers  (2025-10-04T04:14:50Z)
• GrAInS: Gradient-based Attribution for Inference-Time Steering of LLMs and VLMs [56.93583799109029]
  GrAInS is an inference-time steering approach that operates across both language-only and vision-language models and tasks.<n>During inference, GrAInS hidden activations at transformer layers guided by token-level attribution signals, and normalizes activations to preserve representational scale.<n>It consistently outperforms both fine-tuning and existing steering baselines.
  arXiv  Detail & Related papers  (2025-07-24T02:34:13Z)
• Fusion Steering: Prompt-Specific Activation Control [0.0]
  Fusion Steering improves factual accuracy in large language models (LLMs) for question-answering (QA) tasks.<n>This approach introduces flexible steering configurations, including full-layer steering and segmented steering.<n>Under the stricter SimpleQA rubric, segmented steering boosts fully correct responses from 0.0% to 13.1%.
  arXiv  Detail & Related papers  (2025-05-28T16:46:55Z)
• SEAL: Steerable Reasoning Calibration of Large Language Models for Free [58.931194824519935]
  Large Language Models (LLMs) have demonstrated compelling capabilities for complex reasoning tasks via the extended chain-of-thought (CoT) reasoning mechanism.<n>Recent studies reveal substantial redundancy in the CoT reasoning traces, which negatively impacts model performance.<n>We introduce SEAL, a training-free approach that seamlessly calibrates the CoT process, improving accuracy while demonstrating significant efficiency gains.
  arXiv  Detail & Related papers  (2025-04-07T02:42:07Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.