Latest from us - the most promising LLM tutor so far!

I've written a summary of a new paper by Google and Eedi. Their LLM tutor dramatically reduced (although didn't eliminate) hallucinations.

In total there were 5 errors out of 3,617 messages sent by the LLM. Would a human teacher make fewer errors?