# AlignSAE: Concept-Aligned Sparse Autoencoders

arXiv: https://arxiv.org/abs/2512.02004 · HTML: https://mlsanity.com/paper/2512.02004/

## Key facts

- Cited by 4 papers in the map (data through 14 Oct 2026).
- Placed in SAE features and their applications, one of 13 directions in Mechanistic interpretability.
- Builds on 34 of its 49 references that are in the map.
- First posted on arXiv 1 Dec 2025.

## Placed in

- Mechanistic interpretability / [SAE features and their applications](https://mlsanity.com/mechinterp/sae-features-and-their-applications/)

## Builds on (most cited first, 30 of 34)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Finetuned Language Models Are Zero-Shot Learners](https://mlsanity.com/paper/2109.01652/) (2021)
- [Sparse Autoencoders Find Highly Interpretable Features in Language Models](https://mlsanity.com/paper/2309.08600/) (2023)
- [Steering Llama 2 via Contrastive Activation Addition](https://mlsanity.com/paper/2312.06681/) (2023)
- [Concept Bottleneck Models](https://mlsanity.com/paper/2007.04612/) (2020)
- [Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Models](https://mlsanity.com/paper/2403.19647/) (2024)
- [Mechanistic Interpretability for AI Safety -- A Review](https://mlsanity.com/paper/2404.14082/) (2024)
- [Jumping Ahead: Improving Reconstruction Fidelity with JumpReLU Sparse Autoencoders](https://mlsanity.com/paper/2407.14435/) (2024)
- [Physics of Language Models: Part 3.1, Knowledge Storage and Extraction](https://mlsanity.com/paper/2309.14316/) (2023)
- [Llama Scope: Extracting Millions of Features from Llama-3.1-8B with Sparse Autoencoders](https://mlsanity.com/paper/2410.20526/) (2024)
- [Harnessing Deep Neural Networks with Logic Rules](https://mlsanity.com/paper/1603.06318/) (2016)
- [SAEBench: A Comprehensive Benchmark for Sparse Autoencoders in Language Model Interpretability](https://mlsanity.com/paper/2503.09532/) (2025)
- [Steering Language Model Refusal with Sparse Autoencoders](https://mlsanity.com/paper/2411.11296/) (2024)
- [Improving Steering Vectors by Targeting Sparse Autoencoder Features](https://mlsanity.com/paper/2411.02193/) (2024)
- [Sparse Autoencoders Do Not Find Canonical Units of Analysis](https://mlsanity.com/paper/2502.04878/) (2025)
- [Grokked Transformers are Implicit Reasoners: A Mechanistic Journey to the Edge of Generalization](https://mlsanity.com/paper/2405.15071/) (2024)
- [SAEs Are Good for Steering -- If You Select the Right Features](https://mlsanity.com/paper/2505.20063/) (2025)
- [A Survey on Sparse Autoencoders: Interpreting the Internal Mechanisms of Large Language Models](https://mlsanity.com/paper/2503.05613/) (2025)
- [Understanding Neural Networks via Feature Visualization: A survey](https://mlsanity.com/paper/1904.08939/) (2019)
- [Steering Knowledge Selection Behaviours in LLMs via SAE-Based Representation Engineering](https://mlsanity.com/paper/2410.15999/) (2024)
- [Applying sparse autoencoders to unlearn knowledge in language models](https://mlsanity.com/paper/2410.19278/) (2024)
- [Steering Large Language Model Activations in Sparse Spaces](https://mlsanity.com/paper/2503.00177/) (2025)
- [Mechanistic Unlearning: Robust Knowledge Unlearning and Editing via Mechanistic Localization](https://mlsanity.com/paper/2410.12949/) (2024)
- [Rethinking Evaluation of Sparse Autoencoders through the Representation of Polysemous Words](https://mlsanity.com/paper/2501.06254/) (2025)
- [Decoding Dark Matter: Specialized Sparse Autoencoders for Interpreting Rare Concepts in Foundation Models](https://mlsanity.com/paper/2411.00743/) (2024)
- [Feature Hedging: Correlated Features Break Narrow Sparse Autoencoders](https://mlsanity.com/paper/2505.11756/) (2025)
- [How do Transformers Learn Implicit Reasoning?](https://mlsanity.com/paper/2505.23653/) (2025)
- [On the Limits of Sparse Autoencoders: A Theoretical Framework and Reweighted Remedy](https://mlsanity.com/paper/2506.15963/) (2025)
- [Injecting Logical Constraints into Neural Networks via Straight-Through Estimators](https://mlsanity.com/paper/2307.04347/) (2023)
- [SafeSteer: Interpretable Safety Steering with Refusal-Evasion in LLMs](https://mlsanity.com/paper/2506.04250/) (2025)

## Built on by (most cited first, 4 of 4)

- [Interpretable and Steerable Concept Bottleneck Sparse Autoencoders](https://mlsanity.com/paper/2512.10805/) (2025)
- [ReflectRL: Learning from Golden Negative Trajectories via Reflective-to-Direct Reasoning](https://mlsanity.com/paper/2608.03972/) (2026)
- [EchoRL: Reinforcement Learning via Rollout Echoing](https://mlsanity.com/paper/2605.31228/) (2026)
- [Supervised sparse auto-encoders for interpretable and compositional representations](https://mlsanity.com/paper/2602.00924/) (2026)

## Abstract (arXiv)

Large Language Models (LLMs) encode factual knowledge within hidden parametric spaces that are difficult to inspect or control. While Sparse Autoencoders (SAEs) can decompose hidden activations into more fine-grained, interpretable features, they often struggle to reliably align these features with human-defined concepts, resulting in entangled and distributed feature representations. To address this, we introduce AlignSAE, a method that aligns SAE features with a predefined ontology through a "pre-train, then post-train" curriculum. After an initial unsupervised training phase, we apply supervised post-training to bind specific concepts to dedicated latent slots while preserving the remaining capacity for general reconstruction. This separation creates an interpretable interface where specific concepts can be inspected and controlled without interference from unrelated features. Empirical results on GPT-2 across synthetic biographical QA (1-hop, 6 concepts) and compositional relational reasoning (2-hop, 20 concepts) demonstrate that AlignSAE enables precise causal interventions, such as reliable concept swaps (94.8% success in 1-hop), by targeting single, semantically aligned slots. AlignSAE achieves disentanglement improvements of 9--∞× over standard SAEs (measured via Mutual Information Gap), and further supports multi-hop reasoning and a mechanistic probe of grokking-like generalization dynamics. Although our evaluation focuses on GPT-2, AlignSAE itself is model-agnostic.