A forward looking thesis based on "research paper". Deeper in the article it mentions:
> testing keys ranging from 1 to 25 bits—tiny compared to Bitcoin's 256-bit encryption
Paper: https://arxiv.org/pdf/2505.15917
The focus on Bitcoin appears to be synthesis on Decrypt's part - it's not mentioned once in the paper.
20x easier to a target which is 1000x more complex than the systems we have today, and at this stage it's a system which is coherent (as in able to provide answers) for brief intervals, and would require a week per keypair.
I am still skeptical of deep mind. I think I'd like to see non-google critique of their systems behaviour, the algorithmic implications, and the outcomes and projections.