When an AI model doubts evidence that was sound

We gave Claude 70 claims, each with the evidence offered for it. Unguided, it almost never passed weak proof, yet it turned down a third of the sound ones.

Mostly by refusing evidence that was fine. We wrote 70 short cases, each a claim about a Cloudflare site, a payment on the Base chain or a web page, paired with the check someone ran to back it, then had Claude Sonnet and Claude Haiku decide whether the check settled the claim. With no guidance, Sonnet called a flawed check proven once in 45 tries, yet it turned down 9 of the 25 checks that did settle their claim. Haiku never passed a flawed check and turned down 12 of the 25 sound ones. The mistake we had built the test around, a model nodding at a status 200 that proves nothing, was the rare one.

Read the full report on AISkills402: https://aiskills402.com/blog/llm-doubts-sound-evidence