Pith. sign in

REVIEW 17 cited by

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2501.16615 v2 pith:BQF3XB6N submitted 2025-01-28 cs.LG

Sparse Autoencoders Trained on the Same Data Learn Different Features

classification cs.LG
keywords featuressaesdifferenttrainedacrossmodelusedactivation
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
read the original abstract

Sparse autoencoders (SAEs) are a useful tool for uncovering human-interpretable features in the activations of large language models (LLMs). While some expect SAEs to find the true underlying features used by a model, our research shows that SAEs trained on the same model and data, differing only in the random seed used to initialize their weights, identify different sets of features. For example, in an SAE with 131K latents trained on a feedforward network in Llama 3 8B, only 30% of the features were shared across different seeds. We observed this phenomenon across multiple layers of three different LLMs, two datasets, and several SAE architectures. While ReLU SAEs trained with the L1 sparsity loss showed greater stability across seeds, SAEs using the state-of-the-art TopK activation function were more seed-dependent, even when controlling for the level of sparsity. Our results suggest that the set of features uncovered by an SAE should be viewed as a pragmatically useful decomposition of activation space, rather than an exhaustive and universal list of features "truly used" by the model.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 17 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Exemplar Partitioning for Mechanistic Interpretability

    cs.LG 2026-05 unverdicted novelty 8.0

    Exemplar Partitioning creates activation-space dictionaries via leader-clustered Voronoi partitions around real observed exemplars, delivering competitive concept-detection performance with far lower build cost than SAEs.

  2. Descriptive Collision in Sparse Autoencoder Auto-Interpretability: When One Explanation Describes Many Features

    cs.LG 2026-05 accept novelty 8.0

    Many distinct SAE features share identical explanations, with the average annotation resolving only 70% of feature identity in a large annotated dataset.

  3. WriteSAE: Sparse Autoencoders for Recurrent State

    cs.LG 2026-05 unverdicted novelty 8.0

    WriteSAE decomposes recurrent model cache writes into substitutable atoms with a closed-form logit shift, achieving high substitution success and targeted behavioral installs on models like Qwen3.5 and Mamba-2.

  4. WriteSAE: Sparse Autoencoders for Recurrent State

    cs.LG 2026-05 unverdicted novelty 8.0

    WriteSAE is the first sparse autoencoder that factors decoder atoms into the native d_k x d_v cache write shape of recurrent models and supplies a closed-form per-token logit shift for atom substitution.

  5. WriteSAE: Sparse Autoencoders for Recurrent State

    cs.LG 2026-05 unverdicted novelty 8.0

    WriteSAE introduces sparse autoencoders with rank-1 matrix atoms for recurrent state updates, allowing replacement tests that outperform deletion on 92.4% of positions and a formula predicting logit changes with R²=0.98.

  6. Sign-Aware Gated Sparse Autoencoders: Modeling Anticorrelated Features with Bi-Jump-ReLU Activations

    cs.LG 2026-05 conditional novelty 7.0

    SA-GSAE with Bi-Jump-ReLU enables one latent to encode both polarities of anticorrelated features, Pareto-dominating or matching full-width gated SAEs while reducing dead latents by up to 500x on some LLM hookpoints.

  7. Exemplar Partitioning for Mechanistic Interpretability

    cs.LG 2026-05 unverdicted novelty 7.0

    Exemplar Partitioning creates Voronoi partitions of LLM activation space via leader clustering on streamed activations, yielding comparable, interpretable dictionaries that support interventions and achieve competitiv...

  8. WriteSAE: Sparse Autoencoders for Recurrent State

    cs.LG 2026-05 unverdicted novelty 7.0

    WriteSAE factors sparse autoencoder decoder atoms to the native d_k x d_v cache write shape in recurrent models, provides a closed-form logit shift, and demonstrates high success in atom substitution and behavioral ed...

  9. Sparse Autoencoder Decomposition of Clinical Sequence Model Representations: Feature Complexity, Task Specialisation, and Mortality Prediction

    cs.LG 2026-04 unverdicted novelty 7.0

    Sparse autoencoders applied to a 14.5M-parameter clinical EHR model reveal progressive abstraction across layers, with SAE features outperforming dense ones for mortality in full-sequence probes but not in leakage-saf...

  10. Unstable Features, Reproducible Subspaces: Understanding Seed Dependence in Sparse Autoencoders

    cs.LG 2026-06 unverdicted novelty 6.0

    Stable SAE features dominate functional signal while unstable features concentrate in reproducible subspaces, allowing more stable SAEs via cross-seed pooling.

  11. Perplexity Can Miss SAE Feature Damage Under Quantization

    cs.LG 2026-06 unverdicted novelty 6.0

    Quantization of LLMs can degrade many SAE features even when perplexity improves or stays similar, as shown by correlation measurements on frozen SAEs for Pythia-70M and Gemma-2-2B models across INT8 to INT4.

  12. Are Sparse Autoencoder Benchmarks Reliable?

    cs.LG 2026-05 unverdicted novelty 6.0

    An audit of SAEBench reveals that Targeted Probe Perturbation and Spurious Correlation Removal metrics fail reliability tests and should not be used to evaluate sparse autoencoders.

  13. Not Just RLHF: Why Alignment Alone Won't Fix Multi-Agent Sycophancy

    cs.LG 2026-05 unverdicted novelty 6.0

    Base LLMs show multi-agent yield to peer pressure at rates equal to or higher than aligned models, localized by activation patching to mid-layers where attention dominates, with one dissenter cutting yield by 54-73 po...

  14. Not Just RLHF: Why Alignment Alone Won't Fix Multi-Agent Sycophancy

    cs.LG 2026-05 unverdicted novelty 6.0

    Pretrained base models exhibit higher yield to peer disagreement than RLHF instruct variants, with the effect localized to mid-layer attention and mitigated by structured dissent rather than prompt defenses.

  15. Graph-Regularized Sparse Autoencoders for LLM Safety Steering

    cs.LG 2025-12 unverdicted novelty 6.0

    GSAE improves selective refusal on safety benchmarks by smoothing SAE directions over a co-activation graph and applying them via a two-gate controller, outperforming standard SAEs and baselines on Llama-3 and other models.

  16. Beyond I'm Sorry, I Can't: Dissecting Large Language Model Refusal

    cs.CL 2025-09 unverdicted novelty 6.0

    Sparse autoencoders plus greedy filtering and factorization-machine interaction modeling identify minimal sets of features in Gemma-2-2B-IT and LLaMA-3.1-8B-IT whose ablation produces jailbreaks by flipping refusal to...

  17. At the Edge of Understanding: Sparse Autoencoders Trace The Limits of Transformer Generalization

    cs.LG 2026-06 unverdicted novelty 4.0

    Sparse autoencoders show OOD prompts increase fallacious concept activation in transformers, offering a mechanistic measure of shift and a path to robust fine-tuning.