AI's larger working memory creates an audit problem
An essay argues that AI gains leverage from working memory, not mystical reasoning. HN readers asked what happens when the resulting proofs grow beyond any one reader's ability to inspect.
The claim is concrete: a system can keep far more intermediate material active than a person. That makes it able to search many more paths before it commits.
HN's important turn was about verification. A result can be useful while becoming too intricate for a human to audit without structured checkpoints.
The limiting resource is not just an answer
Davide Piffer's essay argues that models can retain much more material than human working memory. In search-heavy technical work, that lets a system keep candidate paths alive rather than choosing early.
The thread shifted to verification
The HN discussion worried about arguments spanning hundreds of books. Another reader described the advantage as persistence: a machine does not tire after a failed research direction.
Build checkable layers
Independent subproofs, testable invariants, and provenance give reviewers places to challenge a result. The bottleneck may be designing reviews that can still reject work too large to hold in one mind.
Get breakdowns like this every day
The top Hacker News stories, unpacked — by email, Telegram, or RSS.
A 232× kernel experiment makes verification the agent skill
A Codex-driven kernel experiment reported a 232× speedup. Its useful lesson is not unsupervised generation: benchmarks, profilers, and a verifier could tell the agent when it was wrong.