A photonic chip plays Mario using a NEAT network trained for just 472 generations
CatAstro_Piyush · x · 2026-11-20
A viral post shows a photonic chip playing Mario, implementing a NEAT (NeuroEvolution of Augmented Topologies) network trained for only 472 generations. The accompanying diagram shows the chip's interferometer layout with phase shifts and splitting ratios per input channel.
The demo highlights an alternative path to neural compute: instead of digital computation on GPUs, optical interference networks handle the forward pass directly in hardware.
More from Fun
- Nous Research Ships Hermes Mobile App, Draws Jokes About a "GTA 6 Integration" — Teknium · 2026-11-20
- Third-party review: Fable model describes what recalling via Mnemos memory feels like — RileyRalmuto · 2026-11-20
- Asking Astra 6.2 to Celebrate an IPO With Maximum Enthusiasm — ravisparikh · 2026-11-20
- Vibe-coded 'Photoshop killers' break on a simple PSD, only Photopea renders it — ssh4net · 2026-11-20
- The internet's pick for best AI video of 2026: 'How the oil was discovered' — CurieuxExplorer · 2026-11-20
- ML community drama: "deep learning is just OLS with constructed regressors" gets meme-rebutted — shakoistsLog · 2026-11-20