Skip to main content


Notes on Claude 3.5 Sonnet (new)'s ability to find errors in Latin text


I took an excerpt from a short story written for beginners, and asked Claude to evaluate it, noting that such short stories often contain errors. In a separate chat, I asked the same question, but replaced "Rōmae" with "Rōmā", which I believe is an error (and Claude in the first chat also thinks is an error). In that chat, Claude also thought the text was correct (but had some unrelated complaints). In a third example, I changed the case of a direct object to the ablative/dative instead of the accusative, and it noticed that. So it looks like Claude is not currently consistent at finding errors in Latin grammar.
in reply to Daniel Filan

altho apparently the "locative ablative" is a thing with names of cities? so maybe Claude is wrong that Rōmā would be wrong?
in reply to Daniel Filan

in reply to Daniel Filan

in reply to Daniel Filan

As for why I care about this capability, see this post


Apparently there is a de facto conspiracy among Latin teachers to not criticize the quality of Latin learning resources they create? Seems not great. foundinantiquity.com/2024/04/1…