The only way to make check whether a LLM output is true is to do the work (to have it dkne by a real person).
For tasks that are trivial to verify, it's ok: a code compiler will run the code written by a LLM. Or: ask a LLM to help you during the examples mapping phase of BDD, and you'll quickly be able to tell what's good and what isn't.
But for the following tasks, there is a risk:
- ask a LLM to make a summary of an email your didn't read. You can't trust the result.
- you're a car mechanic. You dump your thoughts to a voice recorder, and use AI to turn it into a textual structured report. You'd better tripple check the output!
- you're a medical doctor, attempting to do the same trick: you'd have to be extra careful with the result!
And don't count on software testing to make AI tool robust: LLM are non deterministic.
"Microsoft Bing Copilot has falsely described a German journalist as a child molester, an escapee from a psychiatric institution, and a fraudster who preys on widows.
Martin Bernklau, who has served for years as a court reporter [...] asked Microsoft Bing Copilot about himself. He found that Microsoft's AI chatbot had blamed him for crimes he had covered."
That's why I set up my own thing.
I don't care about analytics at all, so I just wrote a simple build system doto generate some very basic HTML redirects.
That was unclear for me too, so I opened an issue (the very first of the project, yay!). The instructions have been updated, they might be a bit clearer now: direct contributions to the code are welcome.
For tasks that are trivial to verify, it's ok: a code compiler will run the code written by a LLM. Or: ask a LLM to help you during the examples mapping phase of BDD, and you'll quickly be able to tell what's good and what isn't.
But for the following tasks, there is a risk: - ask a LLM to make a summary of an email your didn't read. You can't trust the result. - you're a car mechanic. You dump your thoughts to a voice recorder, and use AI to turn it into a textual structured report. You'd better tripple check the output! - you're a medical doctor, attempting to do the same trick: you'd have to be extra careful with the result!
And don't count on software testing to make AI tool robust: LLM are non deterministic.