Program-as-Weights: A Programming Paradigm for Fuzzy Functions

Researchers propose Program-as-Weights (PAW), a new programming paradigm that compiles natural-language specifications into compact neural programs for handling fuzzy tasks like log filtering and JSON repair. A 4B parameter compiler trained on a 10M-example dataset produces efficient adapters that enable a lightweight 0.6B interpreter to match the performance of a much larger 32B language model while using significantly less memory and running locally on consumer hardware. This approach repositions foundation models as function builders invoked once per task definition, making subsequent function calls cheap and reproducible without relying on expensive API services.