Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Social Skill Training with Large Language Models

arxiv.org

Post reply on HN
Print view View original

81–90 of 105 posts

Re: Social Skill Training with Large Language Models

#81

Post by jimbokun »

SamoyedFurFluff wrote: post #79
jimbokun wrote: post #72

Earlier quoted context omitted.

We have run this experiment on a society wide basis. The data are in, and it didn't work. Mental health diagnoses are skyrocketing across the board, especially for young people. And the evidence tracking these outcomes to access to personal phones and social networks for those young people is very strong. Virtual communities have some value. But as a replacement for in person interaction the outcomes are terrible. (B…

Weird take. We know communal lynchings happened; people literally took family photos by them! People happily documented that hangings were the town spectacles when they happened! Let’s not play rose tinted glasses about communities before social media: if you were not “in” that group you were vulnerable to violence or death. We have that now, where a disproportionate number of homeless minors are homeless because the…

Statistically, the mental health of gay kids is worse today than it was when there was a lot more real world persecution.

Re: Social Skill Training with Large Language Models

#82

Post by woopsn »

So, intending no snark here - seminars in confidence and writing, emotional regulation, leadership, language tutoring, anger management, other interpersonal/soft skills.. I imagine a future where these are available to all on tape and CD. LLMs could really make a difference for the effectiveness of "self-help", but the picture that has formed of trust, transparency and ownership in the online data broker industry is pretty ugly. Have you ever heard in any kind of session "what you share in this room will stay in this room"? Or "this is a safe place to make mistakes/say something dumb"? That seems to be effective and necessary part of working with someone else on your growth.

Re: Social Skill Training with Large Language Models

#83

Post by 01HNNWZ0MV43FF »

helboi4 wrote: post #29

Another episode of "what convoluted way can we use tech to 'solve' a social issue that it clearly exacerbates to try and desperately keep the blind cult-like silicon valley version of techno-optimism alive for as long as possible, while ignoring the obvious disastrous social consequences" I'm pretty sure nobody 15 years ago would have uttered the phrase "opportunities for practising social skills are hard to come by"…

> nobody 15 years ago would have uttered the phrase "opportunities for practising social skills are hard to come by"

College was tough for me

Re: Social Skill Training with Large Language Models

#84

Post by jmole »

VeryFinePrint wrote: post #64

As a reddit mod of a tiny community I've very recently started to suggest users use ChatGPT to re-word their comments and submissions when their phrasing is inflammatory. Any point worth making is worth making well. It is still early stages, so I'm not sure how effective it will be. I've noticed that ChatGPT can do a decent job rewording an angry message to be less toxic while drawing out the salient points. The outp…

It'd be hard to read this kind of a response as anything but passive aggressive, unless you are actually sending their comment to ChatGPT on their behalf.

"Please use chatGPT to fix your comment" is just too snide to be useful, like "RTFM" or "just google it".

Re: Social Skill Training with Large Language Models

#86

Post by SJC_Hacker »

ToucanLoucan wrote: post #68
SJC_Hacker wrote: post #57

Earlier quoted context omitted.

I get it, OTOH, humans are unfortunately selfish assholes. In any sufficiently large group of people, there are always conflicts - people attempting to assume control, marginalize/exclude/bully other individuals, etc. Which is part of the reason why people seek out these alternatives. If they ever come out with the sexbot, then I think its all over for modern technological civilization. There will be little reason fo…

> I get it, OTOH, humans are unfortunately selfish assholes. In any sufficiently large group of people, there are always conflicts - people attempting to assume control, marginalize/exclude/bully other individuals, etc. Which is part of the reason why people seek out these alternatives. The prevalence of this is greatly, greatly exaggerated in my experience. School was absolutely like this, but the further you get fr…

> The less anyone givss a shit.

Except when it comes to things like business/jobs. Or romantic partners.

Then people will knife you in the back if they can get an advantage. Or squeeze you as much as possible.

I leave it to you how much of someone's life those two things occupy.

Re: Social Skill Training with Large Language Models

#87

Post by VeryFinePrint »

jmole wrote: post #84
VeryFinePrint wrote: post #64

As a reddit mod of a tiny community I've very recently started to suggest users use ChatGPT to re-word their comments and submissions when their phrasing is inflammatory. Any point worth making is worth making well. It is still early stages, so I'm not sure how effective it will be. I've noticed that ChatGPT can do a decent job rewording an angry message to be less toxic while drawing out the salient points. The outp…

It'd be hard to read this kind of a response as anything but passive aggressive, unless you are actually sending their comment to ChatGPT on their behalf. "Please use chatGPT to fix your comment" is just too snide to be useful, like "RTFM" or "just google it".

I send their comment through ChatGPT, then paste is as an example in the response. "Please use ChatGPT, this is an example of what your comment would look like..."

Re: Social Skill Training with Large Language Models

#88

Post by sandspar »

082349872349872 wrote: post #33
WaitWaitWha wrote: post #21

> However, practice environments for social skills are typically out of reach for most people. I learned my social skills in school, interacting with my peers in my neighborhood, and from my family. I understand this is not available to everyone, but anyone with access the LLMs there is a high likelihood they also (should?) have access to schooling, some neighbors, and family. To what culture this LLM was trained to?

schooling, some neighbours, and family will likely do nothing beyond reflect your own culture back at you. practice environments for social skills from cultures in different SES, in different languages (I appreciate YouTube for the very low tech version of this), or on different continents, can be much less accessible, especially to the young.

Current LLMs also do nothing but reflect your own culture back to you. ChatGPT, Gemini, Claude are American products and reflect an establishment American worldview. Specifically an American culture worldview that's a mix between Silicon Valley, the anxiety prone middle class, and the CIA.

Re: Social Skill Training with Large Language Models

#89

Post by sandspar »

"Social skills practice with AI" is another word for brainwashing, basically. Brainwashing often rotates around ethics, specifically ethics as depicted by social models ("influencers"). And it uses social anxiety as a primary lever (e.g. "assertiveness training" courses, women's deodorant ads). AI "social skills" practice would brainwash susceptible people into mimicking the behavior patterns desired by the creators of the AI, i.e. the ethics desired by the creators. In the case of our current AI's, that behavior pattern/ethic seems to be an amalgamation of performative Silicon Valley "look at how CURIOUS I am!" theatrics and a big business/big government/blob "teamwork with people you don't like" thing.

Re: Social Skill Training with Large Language Models

#90

Post by ben_w »

jimbokun wrote: post #74
ben_w wrote: post #42

Earlier quoted context omitted.

> Think of self-checkouts, chat-bots, AI partners and therapists, technology making us more self-reliant (which is a bad thing). When was the last time you used a bank teller rather than an ATM? For me, at least 5 years ago. On the other hand, despite ATMs, I'm told more people than ever work as bank tellers. > If we keep letting technology take away our communities this way in this two step process (first, take away…

> I'm told more people than ever work as bank tellers. Any links to this? I find this very surprising.

When I first read it, so did I.

Turns out my link was old, it's still more than the 1970s but not at an all time high:

Current: 364,100, https://www.bls.gov/ooh/office-and-administrative-support/te...

(I'm now going to go and update the relevant Wikipedia page…)

Post reply on HN