OpenAI's Automated Interpretability from paper "Language models can explain neurons in language models". Modified by Johnny Lin to add new models/context windows.
the main thing this neuron does is find elements of structured textual data, such as code syntax, URLs, financial figures, or technical specifications, especially when they contain numerical values or specific identifiers.
Text discussing government borrowing, budget deficits or public debt and their macroeconomic effects (financing, inflation, currency/interest-rate consequences).
gpt-5-mini
can cause inflation if financed by money creation, which depreci
The neuron detects numeric content and ratio/proportion notation — i.e., tokens that form numbers, numeric expressions, and colon-separated ratio/proportion formats.
This neuron detects numbers and explicit monetary/financial figures or tokens (amounts, revenues, receipts, taxes, deficits) in financial or budget-related text.
gpt-5-mini
the amount the government takes in and spends rose nearly
This neuron detects dates/years and macroeconomic/financial terms (years, timestamps and words about inflation, interest rates, markets and monetary policy).
gpt-5-mini
: Presidents releasing statements, U.S. increasing interest rates
