6arXiv cs.CL (Computation and Language)·
FEGA framework reveals SAE features encode both concepts and context-dependent functions, complicating steering
A new arXiv preprint introduces Feature-Effect Geometry Analysis (FEGA), an unsupervised framework that studies how sparse autoencoder (SAE) feature interventions affect model logits rather than examining internal feature geometry. The authors find that consistent one-dimensional causal effects are rare across SAE features, and distinguish 'value-like' features (tied to static factual attributes) from 'pointer-like' features (context-dependent operations), with the latter exhibiting diffuse, hard-to-steer effects. The key finding is that a feature can be interpretable and causally relevant without providing a stable steering direction, which has significant implications for mechanistic interpretability workflows that rely on SAE-based steering.
Related guides (1)
Related events (8)
SAEs as Stethoscopes: Interpretability-Guided Layer Selection for Task Vector Model Editing
This paper evaluates a Sparse Autoencoder (SAE)-guided model editing pipeline for mathematical reasoning on Gemma-3-4B-IT, finding that projecting task vectors onto SAE feature subspaces discards ~97% of modification energy due to geometric misalignment between activation-space SAE directions and weight-space task vectors. The authors reframe SAEs as diagnostic tools ('stethoscopes') rather than intervention filters ('scalpels'), using SAE-derived specificity scores to identify which layers to inject unfiltered task vectors into. This approach improves Number Theory accuracy from 29.6% to 39.4% on Minerva Math (p=0.0007), with 5 of 7 math subjects significantly improved and none degraded. The method is fully deterministic and adds no inference cost.
Study finds SAE unstable features reflect reproducible subspaces, not pure noise
A new arXiv paper investigates feature stability in sparse autoencoders (SAEs), measuring the probability that individual learned features reappear across independent training runs. The authors find a functional asymmetry: stable features carry most reconstruction-relevant signal, while unstable features are individually non-reproducible but concentrate in reproducible lower-rank subspaces, suggesting seed dependence reflects basis ambiguity rather than noise. A synthetic model confirms that low-rank ground-truth features can be recovered at the subspace level even when individual SAE latents are non-identifiable across seeds. The work has direct implications for interpretability research that relies on SAE features as meaningful, stable units of analysis.
Procrustes-conditioned Joint SAE extracts cross-seed universal features from BERT models
Researchers introduce a Procrustes-conditioned Joint End-to-end Top-K Sparse Autoencoder (SAE) to address cross-seed feature universality in mechanistic interpretability of BERT models. By applying an orthogonal Procrustes rotation between independently trained models' activation spaces before joint SAE training, the method produces more consistent features (Pearson r ≥ 0.70) than post-hoc alignment baselines across three NLP benchmarks. The work targets a fundamental challenge in dictionary learning: non-convex optimization causes independently trained networks to learn misaligned feature spaces, making it difficult to identify truly universal features. High-universality features are shown to encode interpretable sociolinguistic patterns.
Feature steering via sparse autoencoders reduces look-ahead bias in LLM forecasting
Researchers apply sparse autoencoders to inspect LLM internal states during forecasting tasks, identifying features associated with time-aware versus look-ahead-biased reasoning. Amplifying time-awareness features causally reduces look-ahead bias while preserving general reasoning performance, whereas directly steering look-ahead-bias features has no effect. The work demonstrates that interpretable temporal features can shift LLMs toward more historically grounded forecasting. This is a mechanistic interpretability result with practical implications for LLM-based prediction systems.
SAERL: Using Sparse Autoencoders to Guide LLM Reinforcement Learning Data Engineering
SAERL is a post-training data engineering framework that uses Sparse Autoencoders (SAEs) — a mechanistic interpretability tool — to extract intrinsic model signals for controlling data diversity, difficulty, and quality during RL fine-tuning. The framework applies SAE-space clustering for batch diversity, a difficulty proxy for curriculum ordering, and a quality probe for data filtering. On Qwen2.5-Math-1.5B with GRPO, SAERL achieves 3% average accuracy improvement and reaches target accuracy with 20% fewer training steps. SAE representations transfer across model families and scales, suggesting broad applicability as a lightweight data engineering tool.
VASAE: Vocabulary-Aligned Sparse Autoencoder assigns intrinsic token names to SAE features during training
Researchers introduce VASAE (Vocabulary-Aligned Sparse Autoencoder), a method that trains SAE features with vocabulary-aligned anchoring so each feature is intrinsically named by the nearest token in the model's embedding space. Applied to GPT-2-small and Llama-3.1-8B, VASAE achieves ~90% feature alignment in shallow-to-middle layers without degrading reconstruction quality, though final-layer alignment is limited. The work addresses a longstanding interpretability bottleneck where SAE dictionary features require expensive post-hoc labeling, potentially enabling more scalable mechanistic analysis.
Framework for discovering and controlling personality-like internal representations in LLMs via SAE decomposition
Researchers introduce a framework grounded in Funder's person-situation-behavior triad to identify, intervene on, and validate trait-like internal representations in LLMs. Using sparse autoencoder (SAE) decomposition on contrastive behavior pairs, they locate sparse features corresponding to opposing poles of personality traits, then demonstrate that feature-level interventions produce consistent cross-situational behavioral shifts. Behavioral outcomes on social intelligence tasks show benefit-tradeoff patterns matching human personality research, providing mechanistic evidence that LLMs contain controllable internal structures linking representations to behaviors. The work advances interpretability and controllability of model personality beyond surface-level prompt conditioning.
Sparse AutoEncoder steering reduces Whisper hallucination rate by ~5x without fine-tuning
Researchers investigate hallucination detection and mitigation in OpenAI's Whisper ASR model by probing internal encoder representations. They find that both raw activations and Sparse AutoEncoder (SAE) latents encode linearly separable hallucination signals concentrated in deeper layers. SAE-based activation steering reduces hallucination rates from 72.6% to 14.1% (Whisper small) and 86.9% to 27.3% (Whisper large-v3) on non-speech audio, with minimal WER degradation, approaching fine-tuning-level performance without weight updates.