DeepMind's Gemini Argon Compresses Turing Machines for Busy Beaver, Goldbach, Riemann in Lean 4
robertarail · x · 2026-10-22
Google DeepMind, with Argon and the community, open-sourced the cryptids project: using Gemini Argon to compress small Turing machines encoding major math problems, formally verified end-to-end in Lean 4. Busy Beaver holdout proofs advanced; Goldbach machines compressed 25→21 states, Riemann 120→104, ZF Consistency 393→380. State reductions and full semantics are kernel-verified (machine halts iff the conjecture fails), with 0 sorry from blank tape to Mathlib. Community contributors began improving it within days. Apache-2.0 licensed.
More from Embodied
- Robotics founder defends Voxel: a Unitree G1 wrapper can still be a good investment — carlosdponx · 2026-10-22
- NVIDIA shows Microduck robot duck doing somersaults on RTX Spark after Jensen Huang's request — huggingface · 2026-10-22
- ABB ships RobotStudio HyperReality on NVIDIA Omniverse; Absolute Accuracy calibration is key — lukas_m_ziegler · 2026-10-22
- Tesla Robotaxi logs 3.3M autonomous miles with zero injuries or airbag deployments — aelluswamy · 2026-10-22
- Samsung tests humanoid robot for 24/7 factory shifts, weighing CES 2027 debut — CyberRobooo · 2026-10-22
- A week with the dual-screen iPhone Duo: gestures, apps and the camera compromise — BenBajarin · 2026-10-22