LessWrong user discovers Claude Opus 4.6 makes systematic errors in Ancient Greek exercises, launching public challenge to identify the mistakes.

User began using Claude Opus 4.6 to study Ancient Greek and grade textbook problem sets from Chapter 3

Switched from having the AI grade their work to having it generate answers directly to avoid sycophantic evaluation

Issued an unsupervised challenge asking eligible participants to identify what mistakes Opus 4.6 makes on Ancient Greek fill-in-the-blank exercises

Explicitly excluded people with Ancient or Modern Greek knowledge or native speakers from participating to preserve the challenge's integrity