It’s been widely agreed among psychologists and exasperated members of the general public that conspiracy theorists are beyond salvation by reason alone. Into the mix comes this study suggesting that LLMs are extremely good at persuading people to abandon or weaken their beliefs in conspiracies:
I strongly disbelieve that paper means what people are saying it means. The hype immediately triggers my BS-detector. But it's not worth my time to dig around about details.
Now, regarding "It is so strange hardly anyone on the left tries to be persuasive" - this is a sampling effect. Look at how both patient and skilled a person would need to be, in order to do your recommended persuasion. How many could reasonably qualify? Alas, being on the left does not automatically imbue someone with huge alignment bonuses to Charisma and Wisdom.
Note, it's really not clear that one-on-one persuasion is a very good "helping" strategy for political change. People recommend it as a cliche, but there are a lot of obvious problems in practice. The more "political" left does think about this, and has arguments against putting too much emphasis on personal "therapy".
"I do not think these experimental results reflect a special power of large language models beyond the reach of all or most humans."
Actually I think the LLM might have a particular edge here. It knows just about everything there is to know about the favorite conspiracy of the person it's trying to persuade, probably more than they do, it probably has a good understanding why they believe what they believe and could easily argue the other side with just as much conviction, though its fine tuning disinclines it from doing that. Most sane people are going to stop obsessing over something once they start to believe it's bullshit, but the LLM finds all human language to be equally fascinating and worthy of attention, from the Facebook ramblings of our less mentally well extended family to the very best feats of academics.
> Awkardly the conspiracies tackled included some that seem to me to have merit
Someone should some day do an analysis on Normies doing "analysis" on "conspiracy theorists" (or more accurately: their ironic, dream world propaganda fuelled imagination of them).
> clearly nuts stuff like corporations have a secret cure for AIDS and cancer
Watch out for Naive Realism!
> I do not think these experimental results reflect a special power of large language models beyond the reach of all or most humans. I think many people can persuade perfectly well if they put their mind to it. Even me and you, dear reader.
If humans don't do something fast, I don't think they are ever going to catch up to the power that exists within LLM's today (much of which is not yet identified & harvested).
> I never found persuasion especially difficult.
Would you like to try some on me? You can nominate easy for you topics and I will choose which to use.
> But the results are, I suspect, not unobtainable.
Watch out though: if we do not try, the obtainable cannot be obtained. And as you say: we do not try.
> This is especially true given our great advantage: we can talk to people in real life, not just via text, giving us significant leverage if we use it.
I recommend you check your premises, and your variable types.
> It is so strange hardly anyone on the left tries to be persuasive.
Strange, but not surprising.
> Here is my advice:
It's a great list, but good luck finding a single human who can do even a few of those at a high level, let alone all of them. And if you think about it: why should it be any other way!!?? After all: how many people do you know who can juggle three balls?
> Informal fallacies (FOR GODS SAKE DON’T GO AROUND MENTIONING THEM BY NAME LIKE YOU’RE CASTING SPELLS FROM HARRY POTTER. NO ONE LIKES THE GUY WHO SAYS AD HOMINEM).
This is such good advice!
> The basics of probability theory, including Bayes theorem and its application to everyday life.
And also, what those things almost always are when manifest at the object level (sophisticated hallucination).
Hi, just found your substack by Scott Alexander linking your survey, then came to see.
Commenting here bc 1. you seem to share one of my obsessions, which is bears (I have two gigantic almost life-sized photos of bears as the artwork in my office...love your bear artwork on your substack). Wondering if you like bears because they're cute or terrifying or admirable or all? I have dreams about bears...which are scary dreams where they're attacking me or my pets...at least once a month. Yet I also find them adorable and funny. So I guess my fascination is both attraction and fear...was wondering what yours was.
2. I also have never found persuasion that hard, though so many people claim it's impossible. I know for a fact I have radically changed many people's opinions on politics and religion, and they have told me so. In fact sometimes I have regretted it, as I've talked some people out of their religious beliefs and it seems to have impacted them in a bad way, which was not my intent. Anyway, just wanted to say I agree with you and it certainly can be done, though only if done with genuine good will and the person believes you like them and wants you to like them/admired you at least somewhat. And it also takes time. I think that changing one's mind on a core belief is somewhat like learning a language or perhaps like gaining muscle memory for a skill like playing a musical instrument. At the beginning, they simply don't have the neurons wired in a manner that allows connections between certain words and emotions. They hear certain words and it immediately sets off their neurons flashing in a manner that means BAD and THREAT and SCARY. So it takes time and repeated exposure to rewire things, much like learning a new language. If you have someone in an emotionally open state, like when they fall in love or when they're enjoying humor and laughing, the process can be vastly accelerated.
That sure is asking a lot, both in emotional and time spent on cognitive effort terms, even when it works. In some ways, I find point 8 particular difficult. Going to live with some level of sanity for me means not digging into details about nonsense. At least the other things would lead to useful knowledge or maybe in the end a positive personal relation (seems very optimistic to me, just buying the premise).
Potentially of interest:
https://www.lesswrong.com/posts/hurF9uFGkJYXzpHEE/a-non-magical-explanation-of-jeffrey-epstein
https://www.lesswrong.com/posts/DZoGEHzZNRsMjfpfE/addendum-a-non-magical-explanation-of-jeffrey-epstein
Related, on the topic of “people probably aren’t impervious to reason”: https://slatestarcodex.com/2017/03/24/guided-by-the-beauty-of-our-weapons/
I strongly disbelieve that paper means what people are saying it means. The hype immediately triggers my BS-detector. But it's not worth my time to dig around about details.
Now, regarding "It is so strange hardly anyone on the left tries to be persuasive" - this is a sampling effect. Look at how both patient and skilled a person would need to be, in order to do your recommended persuasion. How many could reasonably qualify? Alas, being on the left does not automatically imbue someone with huge alignment bonuses to Charisma and Wisdom.
Note, it's really not clear that one-on-one persuasion is a very good "helping" strategy for political change. People recommend it as a cliche, but there are a lot of obvious problems in practice. The more "political" left does think about this, and has arguments against putting too much emphasis on personal "therapy".
"I do not think these experimental results reflect a special power of large language models beyond the reach of all or most humans."
Actually I think the LLM might have a particular edge here. It knows just about everything there is to know about the favorite conspiracy of the person it's trying to persuade, probably more than they do, it probably has a good understanding why they believe what they believe and could easily argue the other side with just as much conviction, though its fine tuning disinclines it from doing that. Most sane people are going to stop obsessing over something once they start to believe it's bullshit, but the LLM finds all human language to be equally fascinating and worthy of attention, from the Facebook ramblings of our less mentally well extended family to the very best feats of academics.
> Awkardly the conspiracies tackled included some that seem to me to have merit
Someone should some day do an analysis on Normies doing "analysis" on "conspiracy theorists" (or more accurately: their ironic, dream world propaganda fuelled imagination of them).
> clearly nuts stuff like corporations have a secret cure for AIDS and cancer
Watch out for Naive Realism!
> I do not think these experimental results reflect a special power of large language models beyond the reach of all or most humans. I think many people can persuade perfectly well if they put their mind to it. Even me and you, dear reader.
If humans don't do something fast, I don't think they are ever going to catch up to the power that exists within LLM's today (much of which is not yet identified & harvested).
> I never found persuasion especially difficult.
Would you like to try some on me? You can nominate easy for you topics and I will choose which to use.
> But the results are, I suspect, not unobtainable.
Watch out though: if we do not try, the obtainable cannot be obtained. And as you say: we do not try.
> This is especially true given our great advantage: we can talk to people in real life, not just via text, giving us significant leverage if we use it.
I recommend you check your premises, and your variable types.
> It is so strange hardly anyone on the left tries to be persuasive.
Strange, but not surprising.
> Here is my advice:
It's a great list, but good luck finding a single human who can do even a few of those at a high level, let alone all of them. And if you think about it: why should it be any other way!!?? After all: how many people do you know who can juggle three balls?
> Informal fallacies (FOR GODS SAKE DON’T GO AROUND MENTIONING THEM BY NAME LIKE YOU’RE CASTING SPELLS FROM HARRY POTTER. NO ONE LIKES THE GUY WHO SAYS AD HOMINEM).
This is such good advice!
> The basics of probability theory, including Bayes theorem and its application to everyday life.
And also, what those things almost always are when manifest at the object level (sophisticated hallucination).
Hi, just found your substack by Scott Alexander linking your survey, then came to see.
Commenting here bc 1. you seem to share one of my obsessions, which is bears (I have two gigantic almost life-sized photos of bears as the artwork in my office...love your bear artwork on your substack). Wondering if you like bears because they're cute or terrifying or admirable or all? I have dreams about bears...which are scary dreams where they're attacking me or my pets...at least once a month. Yet I also find them adorable and funny. So I guess my fascination is both attraction and fear...was wondering what yours was.
2. I also have never found persuasion that hard, though so many people claim it's impossible. I know for a fact I have radically changed many people's opinions on politics and religion, and they have told me so. In fact sometimes I have regretted it, as I've talked some people out of their religious beliefs and it seems to have impacted them in a bad way, which was not my intent. Anyway, just wanted to say I agree with you and it certainly can be done, though only if done with genuine good will and the person believes you like them and wants you to like them/admired you at least somewhat. And it also takes time. I think that changing one's mind on a core belief is somewhat like learning a language or perhaps like gaining muscle memory for a skill like playing a musical instrument. At the beginning, they simply don't have the neurons wired in a manner that allows connections between certain words and emotions. They hear certain words and it immediately sets off their neurons flashing in a manner that means BAD and THREAT and SCARY. So it takes time and repeated exposure to rewire things, much like learning a new language. If you have someone in an emotionally open state, like when they fall in love or when they're enjoying humor and laughing, the process can be vastly accelerated.
Nope:
https://www.yudkowsky.net/rational/bayes
https://arbital.com/p/bayes_rule/
That sure is asking a lot, both in emotional and time spent on cognitive effort terms, even when it works. In some ways, I find point 8 particular difficult. Going to live with some level of sanity for me means not digging into details about nonsense. At least the other things would lead to useful knowledge or maybe in the end a positive personal relation (seems very optimistic to me, just buying the premise).