Much more likely is that it would hallucinate a plausible sounding but incorrect answer and send intermediate and junior engineers on a wild goose chase

▲

teravor 9 hours ago | parent | next [-]

if an LLM is capable enough to be used this way it would be used to generate scenarios for the people who would otherwise have to be the ones to generate them. those people would then evaluate the scnearios. those people would then be in a position to decide if the LLM saves them time.

▲

red75prime 11 hours ago | parent | prev [-]

> a plausible sounding but incorrect answer

That is an incorrect but plausible hypothesis. Do you really think that people can't make such mistakes?

If you want to say that people have understanding, then define understanding in an operationalizable way first.

It doesn't mean that I would recommend a general-purpose AI model without additional training to do a fault analysis.

▲

malux85 10 hours ago | parent [-]

>That is an incorrect but plausible hypothesis. >Do you really think that people can't make such mistakes?

Where did I say that? You just pulled that out of nowhere and then refuted it - strawman https://en.wikipedia.org/wiki/Straw_man

▲

red75prime 10 hours ago | parent [-]

> it would hallucinate a plausible sounding but incorrect answer

"Hallucinate" as used in this context does not apply to humans and presupposes a qualitative difference.

▲

malux85 10 hours ago | parent [-]

I was saying that an AI would more likely hallucinate an incorrect answer than correctly diagnose the root cause failure. At no time was I comparing an AI to a human, thats the bit you made up.

▲

Paracompact 9 hours ago | parent | next [-]

The knee-jerk reaction to pointing out any failure modes of AI with, "but meatbags bad!" is a tiring strawman to deal with. It immediately turns the discussion into something else.

	▲	ted_bunny 5 hours ago \| parent [-]
		Humans crash millions of cars a year, and you're worried about one dog driver running over four measly people?

▲

red75prime 9 hours ago | parent | prev [-]

So, your message is "Unspecified AI models with or without additional training aren't ready to do aerospace fault analysis and they can lead experienced engineers astray." OK, it might or mightn't be true depending on the free parameters in your statement.

	▲	malux85 9 hours ago \| parent [-]
		I used the word "likely" meaning there is a chance, your re-phrasing of what I said into a certainty ... and then refuting that certainty, is another textbook strawman argument, you made the same logical fallacy again. Also I said "intermediate and junior" engineers - meaning INexperienced engineers, not experienced ones, so you quoted me wrong in that part too.