The L2 norm of LLM hidden states signals reasoning intensity, with a theoretical bound on SAE feature activations, enabling three new test-time scaling techniques that boost performance.
Controllable llm reasoning via sparse autoencoder-based steering.arXiv preprint arXiv:2601.03595
6 Pith papers cite this work. Polarity classification is still indexing.
6
Pith papers citing it
years
2026 6representative citing papers
Reward-guided optimization of SAE latents, gated by reward and confidence, improves multi-benchmark reasoning and is post-hoc associated with more verification and course correction.
Transcoders decompose MLP layers in Gemma 3-4B-IT to trace visual grounding more effectively than SAEs and predict hallucinations from circuit graph features at AUC 0.68.
Prototype-Based Sparse Steering decomposes query activations with SAEs and optimizes sparse features via gradients to steer LLM outputs toward specific behaviors.
Qwen-Scope provides open-source sparse autoencoders for Qwen models that function as practical interfaces for steering, evaluating, data workflows, and optimizing large language models.
Trains and releases SAEs for Qwen3-1.7B/4B/8B models with layer-wise coverage and demonstrates causal steering of refusal via selected features.
citing papers explorer
-
The Tell-Tale Norm: $\ell_2$ Magnitude as a Signal for Reasoning Dynamics in Large Language Models
The L2 norm of LLM hidden states signals reasoning intensity, with a theoretical bound on SAE feature activations, enabling three new test-time scaling techniques that boost performance.
-
Latent Reward Steering: An Adaptive Inference-Time Framework that Implicitly Promotes Cognitive Behaviors in Reasoning LLMs
Reward-guided optimization of SAE latents, gated by reward and confidence, improves multi-benchmark reasoning and is post-hoc associated with more verification and course correction.
-
Transcoders Trace Visual Grounding and Hallucinations in Vision-Language Models
Transcoders decompose MLP layers in Gemma 3-4B-IT to trace visual grounding more effectively than SAEs and predict hallucinations from circuit graph features at AUC 0.68.
-
Steered Generation via Gradient-Based Optimization on Sparse Query Features
Prototype-Based Sparse Steering decomposes query activations with SAEs and optimizes sparse features via gradients to steer LLM outputs toward specific behaviors.
-
Qwen-Scope: Turning Sparse Features into Development Tools for Large Language Models
Qwen-Scope provides open-source sparse autoencoders for Qwen models that function as practical interfaces for steering, evaluating, data workflows, and optimizing large language models.
-
Discovering Millions of Interpretable Features with Sparse Autoencoders
Trains and releases SAEs for Qwen3-1.7B/4B/8B models with layer-wise coverage and demonstrates causal steering of refusal via selected features.