Gensyn ships open-1b with public proof of its full training process
benfielding · x · 2026-09-15
Gensyn released open-1b, a 1B-parameter language model bundled with something no lab has shipped before: cryptographic proof of how it was trained. A public fingerprint covers every step, and a tool lets anyone replay the training bit-for-bit on their own hardware.
The approach decentralizes verification rather than training itself: initial training stays centralized, avoiding the huge upfront costs and communication issues of decentralized training while keeping the auditability benefits. Gensyn frames open-1b as a milestone toward verifiable AI and a defense against a black-box future.
Related event: Gensyn Releases open-1b, First Open Model with Verifiable Training Proofs(4 posts)→
More from Research
- Odyssey unveils Odyssey-3 foundation world model for robots, humanoids, cars and drones — HeyAmit_ · 2026-09-16
- Johns Hopkins analysis of 1,100+ models claims all deep networks converge to one 16-D subspace — evolvingstuff · 2026-09-16
- EMNLP Paper: Simple OPD-then-RLVR Two-Stage Training Beats All Joint Distillation+RL Baselines — xiye_nlp · 2026-09-16
- OpenAI Foundation commits $125M+ to 'Public Data for Health' science grants — owl_posting · 2026-09-16
- alphaXiv launches Amodei Number, extending Erdős numbers to all AI researchers — burny_tech · 2026-09-16
- AAAI paper: probabilistic hierarchical goal network planning with UCT, compressed search wins at scale — xuanalogue · 2026-09-16