OpenAI's Automated Interpretability from paper "Language models can explain neurons in language models". Modified by Johnny Lin to add new models/context windows.
Default prompts from the main branch, strategy TokenActivationPair.
Recent Explanations
This neuron detects mentions of John Rawls’s work and key terms from his political‐liberal theory (e.g. Rawls, Political Liberalism, justice, fairness, normative principles).
business and finance reporting about specific companies, especially references to corporate performance metrics, stock/share activity, and mergers or acquisitions.
gpt-5
sector.↵↵RIM’s current problems are well documented
language signaling that something has been finished or brought to completion, often in temporal clauses marking “upon/when” completion of an action or process.
numeric identifiers and codes in references/metadata (digits in URLs/DOIs, catalog or gene IDs, phone/patent numbers), with a strong bias toward the sequence 17.
mentions of military installations and unit organizational history, especially bases/airfields, squadrons/wings, and their activations, assignments, movements, and station locations.
gpt-5
military uses Guantanamo Naval base in south Cuba as
special formatting and layout markers—especially unusual whitespace/indentation—that signal tables, section breaks, lists, or math/technical formatting in structured documents.
quoted passages from early modern English records—especially 17th‑century entries with names, dates, and archaic orthography in publication or trial contexts.
gpt-5
to print "A booke called Master William Shakespeare his
