OpenAI lists ten mathematical advances, and HN asks for the denominator
OpenAI says its researchers and systems helped advance ten problems in mathematics and theoretical computer science, with formalization in Lean among the work described.
The announcement presents ten advances and says researchers helped prepare manuscripts and formalize proofs in Lean. That turns a model result into a claim about a research process, not just an answer.
HN’s sharpest question was what the release leaves uncounted: attempted problems, failed runs, repeated attempts, and the cost of the surrounding harness.
Ten results, but not a benchmark card
OpenAI’s account frames the work as ten advances across mathematics and theoretical computer science. It also says the team helped prepare manuscripts and formalize proofs in Lean, a proof assistant whose checker can verify a formal proof.
The number commenters wanted was the failure rate
The 727-comment thread did not mainly dispute that formal verification is useful. It asked what a reader needs to evaluate the reported cost and capability: how many problems were tried, how many attempts each received, what was abandoned, and what compute sat behind the experiment.
That distinction matters because a selected set of successes is not a rate. The thread’s critique is methodological: a research story becomes comparable only when its search budget and its misses are visible.
Lean narrows one question, not all of them
Formalization can make the final proof object checkable, but it does not explain how a conjecture was chosen, how much human guidance was used, or whether the same setup generalizes. The next useful release would separate those layers instead of compressing them into one headline.
Get breakdowns like this every day
The top Hacker News stories, unpacked — by email, Telegram, or RSS.
Retyping AI code is a bet against the comprehension hole
A developer proposes manually retyping LLM-generated code instead of pasting it, trading speed for the small pauses in which a mental model can form.