James TauberGot Sonnet to do some annotation. Told Opus to go fix the errors in the result. I wonder if that's going to turn out to be an effective strategy for large-scale annotation.
Even Sonnet reviewing the annotations of a previous Sonnet run picks up a lot of errors.
Muhammad Shakir محمد شاکر@jtauber.com Used Gemini 2.5 Pro recently for grammatical classification (complement clause versus relative clause). Then reported the classification accuracy by looking at a 2% random sample (manually). That was the best way I could think of to incorporate LLM in the work flow.
what sort of accuracy did you get?