{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/automatically-interpreting-millions-of","title":"Automatically Interpreting Millions of Features in Large Language Models","arxiv_id":"2410.13928","date":"2024-10-17","proceeding":null,"authors":["Gonçalo Paulo","Alex Mallen","Caden Juang","Nora Belrose"],"abstract":"While the activations of neurons in deep neural networks usually do not have a simple human-understandable interpretation, sparse autoencoders (SAEs) can be used to transform these activations into a higher-dimensional latent space which may be more easily interpretable. However, these SAEs can have millions of distinct latent features, making it infeasible for humans to manually interpret each one. In this work, we build an open-source automated pipeline to generate and evaluate natural language explanations for SAE features using LLMs. We test our framework on SAEs of varying sizes, activation functions, and losses, trained on two different open-weight LLMs. We introduce five new techniques to score the quality of explanations that are cheaper to run than the previous state of the art. One of these techniques, intervention scoring, evaluates the interpretability of the effects of intervening on a feature, which we find explains features that are not recalled by existing methods. We propose guidelines for generating better explanations that remain valid for a broader set of activating contexts, and discuss pitfalls with existing scoring techniques. We use our explanations to measure the semantic similarity of independently trained SAEs, and find that SAEs trained on nearby layers of the residual stream are highly similar. Our large-scale analysis confirms that SAE latents are indeed much more interpretable than neurons, even when neurons are sparsified using top-$k$ postprocessing. Our code is available at https://github.com/EleutherAI/sae-auto-interp, and our explanations are available at https://huggingface.co/datasets/EleutherAI/auto_interp_explanations.","url_abs":"https://arxiv.org/abs/2410.13928v2","url_pdf":"https://arxiv.org/pdf/2410.13928v2.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"automatically-interpreting-millions-of","repo_url":"https://github.com/eleutherai/sae-auto-interp","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"Apache-2.0"}}],"tasks":[{"task_slug":"semantic-similarity","task_name":"Semantic Similarity"},{"task_slug":"semantic-textual-similarity","task_name":"Semantic Textual Similarity"}],"methods":[{"method_slug":"set","method_name":"SET"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=2410.13928","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2410.13928"}},"developers":"https://syntology.ai/developers","read_at":"2026-09-24T18:15:14+00:00","read_at_is":"when the build read Syntology's graph, not when any sample ran","claim":"Per-sample execution status on synthesized fixtures; not a correctness claim about the paper. Samples come from repositories linked to the paper, official or community; repo_kind says which.","repos":[{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/eleutherai/sae-auto-interp","reach":{"status":"ok","spdx":"Apache-2.0"}}],"summary":{"ran":3,"unverified":2},"by_repo_kind":{"official":{"samples":5,"ran":3,"repositories":1}},"repo_kind_vocabulary":{"official":"The archive marks this repository official for the paper","named_in_paper":"The archive records that the paper mentions this repository; it is not marked official","listed":"In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper","found_in_text":"Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted","community":"Not in the archive's code links for this paper; a community repository Syntology harvested"},"n_pointer_only_for_licence":0,"samples":[{"code_sha256_prefix":"515e8ab746e6071d","entry":"assert_type","repo":"eleutherai/sae-auto-interp","repo_kind":"official","path":"delphi/utils.py","file_url":"https://github.com/eleutherai/sae-auto-interp/blob/HEAD/delphi/utils.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"515e8ab746e6071d"}},{"code_sha256_prefix":"bdc89f7d6998d834","entry":"non_redundant_hookpoints","repo":"eleutherai/sae-auto-interp","repo_kind":"official","path":"delphi/__main__.py","file_url":"https://github.com/eleutherai/sae-auto-interp/blob/HEAD/delphi/__main__.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"bdc89f7d6998d834"}},{"code_sha256_prefix":"9e0739dc6366e1ee","entry":"process_wrapper","repo":"eleutherai/sae-auto-interp","repo_kind":"official","path":"delphi/pipeline.py","file_url":"https://github.com/eleutherai/sae-auto-interp/blob/HEAD/delphi/pipeline.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"9e0739dc6366e1ee"}},{"code_sha256_prefix":"0b9b30059dc5f9f9","entry":"decode_per_token","repo":"eleutherai/sae-auto-interp","repo_kind":"official","path":"delphi/utils.py","file_url":"https://github.com/eleutherai/sae-auto-interp/blob/HEAD/delphi/utils.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"0b9b30059dc5f9f9"}},{"code_sha256_prefix":"cb784f8270fcbeef","entry":"load_tokenized_data","repo":"eleutherai/sae-auto-interp","repo_kind":"official","path":"delphi/utils.py","file_url":"https://github.com/eleutherai/sae-auto-interp/blob/HEAD/delphi/utils.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"cb784f8270fcbeef"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}