www.lesswrong.com/posts/JANqfGrMyBgcKtGgK/about-30-of-humanity-s-last-exam-chemi...

The cited HLE revision does not say it found that 18% of questions were wrong. It says a targeted peer review of a biology/chemistry/health subset found an expert disagreement rate of approximately 18%. That is a different claim: disagreement among reviewers is not the same as a verified error rate.

This distinction matters because the same passage explains that under a different review rule—flagging a question whenever a single reviewer dissents—the rate rises from 18% to 25%. That shows the figure is about the review methodology and reviewer disagreement, not a direct measurement of how many questions are objectively incorrect.

So the post's paraphrase overstates what Scale AI reported: the source supports an 18% disagreement rate, not an 18% error rate.

The statement overreaches by saying no properties of oganesson have been measured. While its chemical and bulk physical properties are largely unmeasured because so few atoms have been synthesized, researchers have still determined experimental nuclear properties from its observed decay.

For example, authoritative reference sources report a measured/estimated half-life for oganesson-294 (on the order of a millisecond). A half-life is a property of the nuclide. So the accurate claim would be narrower: oganesson’s ordinary physical and chemical properties have not been directly characterized, not that no properties at all have been measured.