# Neural Referee

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Jaewon Jang](https://cerebralvalley.ai/u/Ciportne)
- **GitHub:** https://github.com/pluto2060/neural-referee
- **Demo video:** https://github.com/pluto2060/neural-referee/blob/main/results/video/neural-referee-demo.mp4
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1528

Hardware sometimes computes a wrong value without any error; Meta studied these silent data corruptions across its fleet (Dixit et al., 2021). As computation moves inside models, the same question applies there: when a step goes wrong, which step and which part of the model? Researchers now compile real computers into transformer weights (Percepta's WebAssembly transformer, QwenPU's RISC-V CPU) but only check the final output. Neural Referee is the verifier: Muse Spark 1.3 (Meta Model API) writes the program and its own tests, the program runs inside a 7-layer transformer whose weights are a computer, and the referee bisects hash chains of the execution (about 11 comparisons for a 3,000-step run) and rechecks one step layer by layer. It returns a receipt, or the first wrong step and the faulty sublayer; Muse reads the verdict, re-runs or repairs, and only delivers answers with a passing receipt. Measured on CPU: bit flips in activations or attention memory during execution were located 30 of 30 times, where comparing weight files located 0 of 30; weight corruption 55 of 55; no false alarms on clean programs up to 44,332 steps; one hidden error caught where the final answer was correct but step 1,909 of 3,107 was wrong. Stress test with 8 bit flips per run on 12 tasks: an unchecked pipeline delivered 4 wrong answers; the Muse agent loop delivered 0 wrong (10 correct with receipts, 2 with no compilable program).

## More from Meta Global AI Developer Hackathon

- [Sentra-X](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1525)
- [Altovex CargoLinx](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1526)
- [Keep Talking](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1527)
- [OjO](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1529)
- [MuZip](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1530)
- [InterVibe](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1531)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1528. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt