AI rule summaries are full of errors

[DrunkCat1337]

Yeah this GenAI infection is running deep. It's time premium subscriptions start getting canceled.

A tad dramatic. Rules summary has little to no impact on the game or who you are playing against. If even that insults you you can just leave, you don't have to announce it on every AI post

Has little to no impact. Other than literally the topic of the thread you're in right now. GenAI really melting peoples brains.

[tbhp]

A human created summary, pre-LLM, at least tried in good faith to be accurate with what was stated. Omissions were fine as you'd just go to the book then, but misinformation is bad. Better to not state anything than to mislead that 2+2=5.

Is it true that AI intentionally does bad faith, or is it just that it's oblivious to its own misunderstandings?

I think the latter is more likely unless it's been specifically programmed to mimic human behavior, in which case it may exhibit bad faith sometimes because that's what humans do.

That, or the person who programmed it just forgot to include "I don't know" as an option. So the AI will always try to come up with some answer.

[lefalaf]

A human created summary, pre-LLM, at least tried in good faith to be accurate with what was stated. Omissions were fine as you'd just go to the book then, but misinformation is bad. Better to not state anything than to mislead that 2+2=5.

Is it true that AI intentionally does bad faith, or is it just that it's oblivious to its own misunderstandings?

I think the latter is more likely unless it's been specifically programmed to mimic human behavior, in which case it may exhibit bad faith sometimes because that's what humans do.

That, or the person who programmed it just forgot to include "I don't know" as an option. So the AI will always try to come up with some answer.

Misinformation (no intent, neutral), not disinformation (malicious intent, bad faith) as LLMs do not have an intelligence to create good or bad intent. It's just probabilities and statistics to create relations between things.

From what I read, they do seem to be programmed to be confident whatever the result, but will back down if called out. Seems they don't know what they don't know (no actual intelligence after all), and most, if not all, probably have a bias towards giving any, or nearly any, result, even if low probability of accuracy.

[RobertBr]

Is it true that AI intentionally does bad faith, or is it just that it's oblivious to its own misunderstandings?

That is a deep philosophical question involving what is meant by 'intention' and behaviouralism. GenAI training process and its commercial success depends in part on its ability to 'please' its audience which almost inevitably means providing an answer (human beings have a strongly inbuilt preference for a wrong answer over no answer). Since that is what generates a lot of its 'hallucinations' you could choose to see it as intentional bad faith.

[lefalaf]

intentional bad faith.

On a more meta level, the people promoting LLMs as intelligent, capable, trustworthy, etc., and suitable to replace humans are very much acting in bad faith.

Even if they were capable, the mass layoffs, environmental harms, and resource demands aren't worth it.