# Discovering and Steering Interpretable Concepts in Large Generative Music Models

arXiv: https://arxiv.org/abs/2505.18186 · HTML: https://mlsanity.com/paper/2505.18186/

## Key facts

- Cited by 13 papers in the map (data through 14 Oct 2026).
- Placed in Activation steering methods, one of 13 directions in Mechanistic interpretability.
- Builds on 19 of its 43 references that are in the map.
- First posted on arXiv 18 May 2025.

## Placed in

- Mechanistic interpretability / [Activation steering methods](https://mlsanity.com/mechinterp/activation-steering-methods/)

## Builds on (most cited first, 19 of 19)

- [WaveNet: A Generative Model for Raw Audio](https://mlsanity.com/paper/1609.03499/) (2016)
- [Concept Bottleneck Models](https://mlsanity.com/paper/2007.04612/) (2020)
- [Large-scale Contrastive Language-Audio Pretraining with Feature Fusion and Keyword-to-Caption Augmentation](https://mlsanity.com/paper/2211.06687/) (2022)
- [AudioLDM: Text-to-Audio Generation with Latent Diffusion Models](https://mlsanity.com/paper/2301.12503/) (2023)
- [Scaling and evaluating sparse autoencoders](https://mlsanity.com/paper/2406.04093/) (2024)
- [Simple and Controllable Music Generation](https://mlsanity.com/paper/2306.05284/) (2023)
- [MusicLM: Generating Music From Text](https://mlsanity.com/paper/2301.11325/) (2023)
- [Make-An-Audio: Text-To-Audio Generation with Prompt-Enhanced Diffusion Models](https://mlsanity.com/paper/2301.12661/) (2023)
- [AxBench: Steering LLMs? Even Simple Baselines Outperform Sparse Autoencoders](https://mlsanity.com/paper/2501.17148/) (2025)
- [k-Sparse Autoencoders](https://mlsanity.com/paper/1312.5663/) (2013)
- [FMA: A Dataset For Music Analysis](https://mlsanity.com/paper/1612.01840/) (2016)
- [SampleRNN: An Unconditional End-to-End Neural Audio Generation Model](https://mlsanity.com/paper/1612.07837/) (2016)
- [Mustango: Toward Controllable Text-to-Music Generation](https://mlsanity.com/paper/2311.08355/) (2023)
- [Discover-then-Name: Task-Agnostic Concept Bottlenecks via Automated Concept Discovery](https://mlsanity.com/paper/2407.14499/) (2024)
- [InterPLM: Discovering Interpretable Features in Protein Language Models via Sparse Autoencoders](https://mlsanity.com/paper/2412.12101/) (2024)
- [DecoderLens: Layerwise Interpretation of Encoder-Decoder Transformers](https://mlsanity.com/paper/2310.03686/) (2023)
- [Do Music Generation Models Encode Music Theory?](https://mlsanity.com/paper/2410.00872/) (2024)
- [TensorFlow Audio Models in Essentia](https://mlsanity.com/paper/2003.07393/) (2020)
- [Melody-Guided Music Generation](https://mlsanity.com/paper/2409.20196/) (2024)

## Built on by (most cited first, 13 of 13)

- [AudioSAE: Towards Understanding of Audio-Processing Models with Sparse AutoEncoders](https://mlsanity.com/paper/2602.05027/) (2026)
- [TADA! Tuning Audio Diffusion Models through Activation Steering](https://mlsanity.com/paper/2602.11910/) (2026)
- [Whisper Hallucination Detection and Mitigation via Hidden Representation Steering and Sparse AutoEncoders](https://mlsanity.com/paper/2606.07473/) (2026)
- [Sparse Autoencoders Make Audio Foundation Models more Explainable](https://mlsanity.com/paper/2509.24793/) (2025)
- [Probing Perceptual Priors of MLLMs via Gibbs Sampling with Interpretable Generative Controls](https://mlsanity.com/paper/2608.27727/) (2026)
- [MusGU+: Toward a Musician-Centered Evaluation Framework and Discovery Tool for Generative Music AI](https://mlsanity.com/paper/2608.30940/) (2026)
- [A Quantized Native Runtime for On-Device Semantic Audio Generation](https://mlsanity.com/paper/2607.08526/) (2026)
- [MI-MIDI: Mechanistic Interpretability of Text-to-MIDI Generation Models via Probing, Lenses and Steering](https://mlsanity.com/paper/2608.06638/) (2026)
- [What Makes a Good Layer? Assessing the Layer-Wise Intrinsic Properties of Music Foundation Models](https://mlsanity.com/paper/2608.14819/) (2026)
- [Pitch-class Steering for Diffusion-based Music Generation via Latent-space Probes](https://mlsanity.com/paper/2609.04516/) (2026)
- [Ghost in the Encoder: Decodable Artist Identity Representations in Lyrics-to-Song Generation](https://mlsanity.com/paper/2609.39552/) (2026)
- [From Isolated Feature to Orbits: Discovering Music Concepts via Multi-SAE Alignment](https://mlsanity.com/paper/2610.01864/) (2026)
- [Harmonic Representations in Symbolic Music Transformers are Localized and Controllable](https://mlsanity.com/paper/2610.15957/) (2026)

## Abstract (arXiv)

The fidelity with which neural networks can now generate content such as music presents a scientific opportunity: these systems appear to have learned implicit theories of such content's structure through statistical learning alone. This offers a potentially new lens on theories of human-generated media. When internal representations align with traditional constructs (e.g. chord progressions in music), they show how such categories can emerge from statistical regularities; when they diverge, they expose limits of existing frameworks and patterns we may have overlooked but that nonetheless carry explanatory power. In this paper, focusing on autoregressive music generators, we introduce a method for discovering interpretable concepts using sparse autoencoders (SAEs), extracting interpretable features from the residual stream of a transformer model. We make this approach scalable and evaluable using automated labeling and validation pipelines. Our results reveal both familiar musical concepts and coherent but uncodified patterns lacking clear counterparts in theory or language. As an extension, we show such concepts can be used to steer model generations. Beyond improving model transparency, our work provides an empirical tool for uncovering organizing principles that have eluded traditional methods of analysis and synthesis.