Show HN: Neuronpedia, an open source platform for AI interpretability
a year ago
- #machine learning
- #AI interpretability
- #open source
- Neuronpedia is now open source as of March 2025.
- It offers 4TB of interpretability data for AI models.
- The platform supports probes, latents/features, custom vectors, and concepts.
- Recent releases include SAEs (Sparse Autoencoders) for models like Llama-3.1-8B and GPT2-Small.
- Supported models include DeepSeek-R1-Llama-8B, Gemma-2-2B-IT, and GPT2-Small.
- Neuronpedia was created by Johnny Lin, an ex-Apple engineer, and is supported by Decode Research and the Long Term Future Fund.