Voices
Willison's Read on the Mythos Cryptanalysis: The Reusable Artifact Is the Prompt, Not the Attack
Simon Willison's July 28 writeup argues the best part of Anthropic's cryptanalysis paper is the published prompts, quoting the researchers directly: 'the models tend to think it is impossible to solve so they don't try they need a good amount of prompting.' He notes the main human interventions across the 60-hour run were to encourage the model not to give up and to 'find something that's worth publishing.' For anyone running long-horizon agents, that is the transferable lesson — the failure mode on genuinely hard problems is premature abandonment, and the fix is a persistence instruction, not a bigger model.
↳ Follow the thread