IA-002 / 2023
Towards Monosemanticity is highlighted textOrdinary text with words shaded in, where the shading shows how strongly the model reacted to each word.Running text with individual words or tokens shaded to encode a per-token value, most often how strongly a feature activated there.glossary showing featuresA single thing the model has learned to recognise, like a place, a tone of voice, or a kind of mistake.Individual learned directions or concepts inside the model.glossary, about superpositionA model packing more ideas into its wiring than it has room for, by letting them overlap.Representing more features than there are dimensions, by encoding them sparsely and near-orthogonally.glossary, polysemanticityOne part of the model doing several unrelated jobs at once.A single component responding to several unrelated concepts.glossary and feature-interpretationWorking out what one small piece of the model has learned to recognise.What an individual learned feature means.glossary.
- ID
- IA-002
- Name
- Towards Monosemanticity — feature browser interface
- Artefact type
- Interactive article
- Visual form
- Highlighted textOrdinary text with words shaded in, where the shading shows how strongly the model reacted to each word.Running text with individual words or tokens shaded to encode a per-token value, most often how strongly a feature activated there.glossary
- Makes visible
- FeaturesA single thing the model has learned to recognise, like a place, a tone of voice, or a kind of mistake.Individual learned directions or concepts inside the model.glossary
- Role of the visual
- Exhibit
- Runnable by a visitor
- Pre-rendered, view only
- Date published
- 2023-10-05
- Year
- 2023
- Link status
- Live
- Built by (people)
- Trenton Bricken; Adly Templeton; Joshua Batson; Brian Chen; Adam Jermyn; Tom Conerly; Nicholas L. Turner; Cem Anil; Carson Denison; Amanda Askell; et al.
- Organisation
- Anthropic
- Venue / published in
- Transformer Circuits Thread
- Sector
- Frontier lab
- Country / region
- US
- Open source
- No
- Method / technique
- Sparse dictionary learning / sparse autoencoders (SAE) over a one-layer transformer's MLP activations
- Model(s) studied
- One-layer transformer (512 neurons); SAE dictionaries ~2k-100k features
- Access required
- Internal / proprietary
- What it visualises
- Individual SAE features: top activating dataset examples, activation histograms, and the logit effects each feature has on the next token
- Interaction affordances
- Browse a feature index; click through to per-feature pages; compare a real neuron against a learned feature side by side
- Pragmatic vs Basic science
- Basic science
- Reverse-eng vs Concept-based
- Reverse-engineering
- Observational vs Interventional
- Observational
- Intended audience
- Researchers
- Description (card)
- The paper that made sparse autoencoders the field's dominant method, published with a browsable interface over every extracted feature so readers could check the monosemanticity claim themselves rather than take the authors' word for it.
- Why it matters
- Turned a contested claim (features are more interpretable than neurons) into something a reader could audit by clicking.
- Visual / design notes
- Dense, utilitarian, research-grade. Feature pages are information-rich rather than designed. The contrast with Distill's polish is itself informative.
- Tags
- modality:languageText. Models that read and write words.Models that generate or process text.glossarymethod:sparse-autoencoderA technique for pulling a model's tangled internals apart into separate, nameable pieces.Learning an overcomplete, sparsely activating basis for a layer's activations.glossaryphenomenon:superpositionA model packing more ideas into its wiring than it has room for, by letting them overlap.Representing more features than there are dimensions, by encoding them sparsely and near-orthogonally.glossaryphenomenon:polysemanticityOne part of the model doing several unrelated jobs at once.A single component responding to several unrelated concepts.glossaryphenomenon:feature-interpretationWorking out what one small piece of the model has learned to recognise.What an individual learned feature means.glossary
- Citation
- Bricken, T., Templeton, A., Batson, J., et al., 2023. Towards Monosemanticity: Decomposing Language Models With Dictionary Learning. Transformer Circuits Thread.
- Related entries
- IA-003
- Confidence
- High
- Source of info
- Read the article and its setup/interface section, 2026-08-06
- Date added
- 2026-08-06
- Added by
- Claude
- Notes
- Author list truncated — full list on the paper. Scott Alexander's ACX piece is a good lay companion.