Great points and nice write-up.

At the risk of sounding AI pilled, I find LLM's are fairly good sparring partners if you instruct them well. At least they're more capable of giving you the best arguments against your position than the average bar patron.

Almost all of my non-professional prompts are of that nature.

▲

Lord-Jobo a day ago | parent | next [-]

LLM are so insanely terrible at this that I constantly have to moderate my prompts with “maybe this isn’t true” or “maybe I am misunderstanding this explanation” to get them to STOP crumbling into dust over the slightest pressure. That’s WITH strong pre prompts that also push against this bias. Roleplay/persona used to be a good way to get around this. “Pretend you are an almighty, rarely wrong, curmudgeonly professor with 150 different phds talking to me, a student”. But these abilities have been strongly restricted due to other (fair) issues so it’s not a solution any more.

Better than a bar patron, no doubt. Still nothing compared to any normal well educated human being

▲

kenferry 3 days ago | parent | prev [-]

I don't know. AI are basically trained to tell you what you want to hear, including when they're disagreeing with you. If you come off of your conversation with AI with a stronger belief than when you went in, is that merited?

"AI Advice Suppresses People’s Willingness To Say “I Don’t Know”, Even When The Advice Is Wrong And Accuracy Is Incentivized" https://osf.io/preprints/psyarxiv/5y6m4_v1

	▲	childintime 2 days ago \| parent [-]
		Yeah you have to frame the prompt as if you've come across some potentially extremist view you want to know more about, and pretend you're an outsider needing advice. Then you bias it to favor what you really want to know.