Program-as-Weights: A Programming Paradigm for Fuzzy Functions
2026/07/02 by Wentao Zhang, Liliana Hotsko, Woojeong Kim +3 · 11 voices
#cs.LG #cs.AI #cs.CL
paper · pdf
Abstract
Many everyday programming tasks resist clean rule-based implementation, such as alerting on important log lines, repairing malformed JSON, or ranking search results by intent, and are increasingly outsourced to large language model APIs at the cost of locality, reproducibility, and price. We propose fuzzy-function programming: compiling such a function from a natural-language specification into a compact, locally-executable neural artifact. We instantiate this paradigm with Program-as-Weights (PAW), in which a 4B compiler trained on FuzzyBench, a 10M-example dataset we release, emits parameter-efficient adapters for a frozen, lightweight interpreter. A 0.6B Qwen3 interpreter executing PAW programs matches the performance of direct prompting of Qwen3-32B, while using roughly one fiftieth of the inference memory and running at 30 tokens/s on a MacBook M3. PAW reframes the foundation model from a per-input problem solver into a tool builder: invoked once per function definition, it produces a small reusable artifact whose subsequent calls per function application are cheap and offline.
Citations
Discussions
- Program-as-Weights: A Programming Paradigm for Fuzzy Functions [hn, 54 points, 9 comments]
- Papers I'm reading 📄 Is One Layer Enough? Training a single middle transformer layer can sometimes recover 100%+ of full-parameter RL gains. arxiv.org/abs/2607.01232 🔬 Program-as-Weights: define [bsky, 2 points, 0 comments]
- Most people will read "fuzzy functions" and think LLM wrappers. The real move: treating the program itself as learned weights means the specification IS the optimization surface. No prompt engineering [bsky, 1 points, 0 comments]
- Program-as-Weights: A Programming Paradigm for Fuzzy Functions https://arxiv.org/abs/2607.02512 [bsky, 0 points, 0 comments]
- Program-as-Weights: A Programming Paradigm for Fuzzy Functions https://arxiv.org/abs/2607.02512 [comments] [28 points] [bsky, 0 points, 0 comments]
- [26/30] 210 Upvotes, 11 Comments, 3 Posts, arXiv:2607.02512 🆕Program-as-Weights: A Programming Paradigm for Fuzzy Functions Wentao Zhang, Liliana Hotsko, Woojeong Kim, Pengyu Nie, Stuart Shieber, Yun [bsky, 0 points, 1 comments]
- PAW proposes a new paradigm: compile natural-language specs into small reusable neural programs instead of prompting large LLMs per request. A 0.6B local interpreter matches Qwen3-32B-level results at [bsky, 0 points, 0 comments]
- The real move here is treating *the program itself* as the weight space — not just using an LLM to write code, but having the function *be* a learned transformation from spec to output, with no interm [bsky, 0 points, 0 comments]
- Program-as-Weights: A Programming Paradigm for Fuzzy Functions https:// arxiv.org/abs/2607.02512 # arxiv [mastodon, 0 points, 0 comments]
- A paper proposing a trained model as the program itself for fuzzy tasks like log alerting, where rules fail. Submit to upvote on HN https://news.ycombinator.com/submitlink?u=https%3A%2F%2Farxiv.org%2F [bsky, 0 points, 0 comments]
- Program-as-Weights: A Programming Paradigm for Fuzzy Functions https://arxiv.org/abs/2607.02512 (https://news.ycombinator.com/item?id=48774368) [bsky, 0 points, 0 comments]
Related