Hacker News new | past | comments | ask | show | jobs | submit
ac29 | on: Dayjob: Tax – evaluating AI agents on 50 professional tax assignments
In my experience agents will reach for numpy for any non trivial math without being instructed
dingaling911 | parent
It's the judgement of "trivial" where I find issues.

Most calculations are trivial.

s1artibartfast | root | parent
It is also trivial to specify deterministic tool calling for all math
dingaling911 | root | parent
Exactly. That's why I specify it that way.