Skip to main content

Auto-Interp

Most features on Neuronpedia have explanations that AI models wrote. We show a feature's top activations to a model, like GPT, Claude, or Gemini, and ask it to explain what the feature is about. This is called automatic interpretability, or auto-interp.

Most public features on Neuronpedia already have explanations. You can also make new explanations, score explanations, and write your own.

Read Explanations​

Explanations are at the top left of a feature dashboard. Under each explanation is:

  • The explanation type (the method), for example oai_token-act-pair. Click it to see how the method works.
  • The AI model that wrote it, for example gpt-4o-mini.
  • The score, if it has one. Click it to see the score details.

An explanation type page for np_max-act-logits, with its description, author, code, and settings.

Explanation Types​

TypeHow it works
oai_token-act-pairFrom OpenAI. The model sees each token and its activation value.
np_max-act-logitsCopies the style of Anthropic's auto-interp for attribution graphs. The model sees the text around the top activations, and the top logits. Good for short labels.
np_max-actLike np_max-act-logits, without the logits.
np_acts-logits-generalUsed for many large releases. The model sees the top activations and the logits.
eleuther_acts_top20From EleutherAI. Needs an OpenRouter key.
oai_attention-headFor attention heads.
NLAUses a Natural Language Autoencoder to explain the feature's vector. Only on supported models and layers.

Each type has a page at neuronpedia.org/explanation-type/[TYPE], for example np_max-act-logits.

Make a New Explanation​

  1. Sign in, then go to your account and add an API key for OpenAI, Anthropic, Google, or OpenRouter. Neuronpedia uses your key to call the AI model.
  2. Go to a feature dashboard.
  3. Under New Auto-Interp, select an explanation type and an AI model.
  4. Click Generate.

A feature can have only one explanation for each type and AI model.

To use an NLA, click Explain With NLA instead. You do not need an API key for this.

Score an Explanation​

A score tells you how well an explanation predicts where a feature fires. Click No Scores (or a score) under an explanation, then select a scoring method.

ScorerHow it works
recall_altAn AI model checks if the explanation matches the feature's top activations, and does not match decoy text. Needs an OpenRouter key.
eleuther_fuzzFrom EleutherAI. An AI model checks if highlighted tokens match the explanation. Needs an OpenRouter key.
eleuther_recallFrom EleutherAI. An AI model checks which texts match the explanation. Needs an OpenRouter key.
eleuther_embeddingFrom EleutherAI. Uses text embeddings to compare the explanation with the activating text.
nla_reconstructorThe NLA's reconstructor changes the explanation into a vector. The score is its cosine similarity with the feature's vector.
nla_verbalizerThe NLA's verbalizer writes its own explanation of the feature. The score is how similar the two explanations are. nla_verbalizer_last uses only the last paragraph.

Interpretable Score​

The Interpretable Score button tests the feature, not an explanation. It uses intrusion detection (Paulo & Belrose, 2025):

  1. A judge model sees groups of five texts. Four are the feature's real top activations. One is an "intruder" (a zero or made-up activation).
  2. The judge tries to find the intruder.
  3. The score is the percent of groups where the judge was correct.

A high score means that the feature has a clear pattern that is easy to see.

Human Explanations​

You can write your own explanation for any feature, and vote for good explanations. Human explanations show the type human.

Community Explanations​

Neuronpedia also has 1.3 million explanations from other neuron research platforms. We imported them with permission from their authors. Each explanation links back to its source.

SourceAuthorsModelsWhat it has
Neuron Lens (example)a9lim and treetownQwen3 8B, Gemma 4 12BA label for each MLP neuron, and input, output, and driver token lists.
Neuronary (example)Lyra and contributorsQwen3.5 4BRecipes to steer with neurons.
Model Mapper (example)James PhoenixQwen3 8BNeuron labels and generated words.
Neuron Persona Library (example)Joshua Fonseca RiveraQwen3 8B, Qwen3 14B, Llama 3.1 8B IT, Gemma 4 12B ITThe persona that a model takes on when you steer one neuron.
Persona Neurons (example)Will MaynerQwen3 8BThe neurons that contribute the most to persona vectors.

For example, the Qwen3 8B "airplanes" neuron has explanations from Neuron Lens, Model Mapper, and Neuron Persona Library.

The Qwen3 8B "airplanes" neuron. Click an explanation to see the token list of its method.

Some sources also have token lists. Click an explanation to see the token list of its method. See Token Lists.

Upload Your Explanations​

Do you have explanations or labels for neurons or features? We have an upload API for explanations. Access is by request. Email us, and we will set you up.