>>106331314
>And yet LLMs will still fail at simple calculations that aren't in the training data
not true for a while.
was a 60% failure rate so couldn't have been a tool call.