DeepMind Factors 68-Digit Semiprimes with Discrete Diffusion Models

DeepMind researchers including Demis Hassabis and Volodymyr Mnih published a preprint showing that Transformers trained with discrete denoising diffusion and guided beam-search decoding can factor 68-digit semiprimes, setting a new record, though RSA encryption remains unaffected.

2026-11-10 ~ 2026-11-10 · 3 related posts

• DeepMind paper: diffusion model factors 68-bit semiprimes, but compute costs scale exponentially — weballergy · 2026-11-10
• New DeepMind paper lands with 13-author lineup incl. Hassabis and Mnih — weballergy · 2026-11-10

1 near-duplicate retellings: weballergy