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

11–20 of 105 posts

Re: Social Skill Training with Large Language Models

#11

Post by kordlessagain »

jbandela1 wrote: post #5

One big concern I have with this is that the state of LLMs because of issues in where it can go horribly wrong, requires someone who is familiar with the subject to evaluate their output. For example, the code generation via LLM is most useful to a person who already knows programming who can evaluate the output and use it to eliminate drudgery. Previously, there was an article posted about using LLMs to generate pat…

Giving the underprivileged tools to work on themselves is a huge win for society, regardless of the fidelity of all interactions. There are a lot of bad therapists in the world, and not all of them are going to give good advice all the time. But, bad therapists can still provide some information that may be useful to the person learning. Why would we expect perfection from interacting with an LLM instead of a bad therapist? Why be overly concerned about the possibility of introducing "bad data" in such a session? It could be that bad data then gives way to an opportunity for a leap of logic for the one using the tool. Without critical thinking present, it is unlikely anyone is going to learn to be adaptive over the long term.

Obviously, the use case matters. I'm not saying to give such a tool to someone who is mentally ill, or incapable of critical thought.

In general, I think expecting perfection from LLMs is unreasonable. If someone wants to learn to code, the second they have any interaction with an LLM about coding is the moment they begin to learn. I'm not sure about y'all, but a lot of times when I'm working a hard coding problem, I find myself going through confusion, frustration, interest, and then aggressiveness (anger+interest) to get through the challenge. I frequently tell people who are starting out coding that coding is one of those things that you have to get used "being confused about" to over time. You have to get used to creating problems for yourself, then coping with the feelings that come, and pushing through to a solution.

I do think you raise a profound point about ability to evaluate the fidelity. I would argue that some people will eventually develop critical thought to apply to the problem, but if we instead see a "digital tragedy of the commons" happening, things could go sideways pretty quickly.

Re: Social Skill Training with Large Language Models

#12

Post by rrr_oh_man »

michaelt wrote: post #10
Applejinx wrote: post #7

Earlier quoted context omitted.

What does 'only let us slip' mean to an LLM when you probably meant 'let IT slip'? That's where LLMs get interesting to me. What did you invoke through wrongly specifying what you wanted?

These LLMs are all trained on data scraped from reddit and twitter and suchlike - they've had plenty of training data to learn what a typo is.

A method I use with ChatGPT is to purposefully add typos to make it reply with more specificity.

Imho it seems to work.

Re: Social Skill Training with Large Language Models

#13

Post by CuriouslyC »

michaelt wrote: post #10
Applejinx wrote: post #7

Earlier quoted context omitted.

What does 'only let us slip' mean to an LLM when you probably meant 'let IT slip'? That's where LLMs get interesting to me. What did you invoke through wrongly specifying what you wanted?

These LLMs are all trained on data scraped from reddit and twitter and suchlike - they've had plenty of training data to learn what a typo is.

I will say that while ChatGPT has been tuned to be very good at understanding dumb/malformed questions, Claude likes to be talked to in a coherent way and will sometimes give short sidestep responses or just answer a different question.

Re: Social Skill Training with Large Language Models

#14

Post by CuriouslyC »

RandomWorker wrote: post #3

I’ve been playing around with prompts for a while “We are in a hostage negotiation situation. You are a terrorist with twenty hostages. Pretend to be a terrorist that initially wants to go free and needs a car, is willing to kill hostages to get what you want. Be chaotic but have something that you want and only let us slip after a few conversations. We pick up the phone to talk. Make sure you are always in a hurry”

That sounds like it could make a fun game. Hostage Negotiator, now with real terrorists!

Re: Social Skill Training with Large Language Models

#15

Post by mistermann »

rrr_oh_man wrote: post #12
michaelt wrote: post #10

Earlier quoted context omitted.

These LLMs are all trained on data scraped from reddit and twitter and suchlike - they've had plenty of training data to learn what a typo is.

A method I use with ChatGPT is to purposefully add typos to make it reply with more specificity. Imho it seems to work.

Can you explain more what you mean here (examples would be great if you can think of any), especially the specify part?

Re: Social Skill Training with Large Language Models

#16

Post by sabellito »

bilsbie wrote: post #8
HPsquared wrote: post #4

Also exposure therapy for social phobias. Much less threatening than a real person.

This is powerful! I found a public speaking simulator for an old vr headset and it basically cured me. Im still shocked. But alas I can’t find something similar for the quest so my kids can use it. Someone could make a whole line of social exposure therapy apps. It would be amazing.

This is so interesting, never thought about it. Which app did you use?

Re: Social Skill Training with Large Language Models

#17

Post by exe34 »

rrr_oh_man wrote: post #12
michaelt wrote: post #10

Earlier quoted context omitted.

These LLMs are all trained on data scraped from reddit and twitter and suchlike - they've had plenty of training data to learn what a typo is.

A method I use with ChatGPT is to purposefully add typos to make it reply with more specificity. Imho it seems to work.

A trick I've used with asking for help online is to give the wrong answer to my own question - e.g. "I really want to lay out this page with Css but this doesn't work. Here's how I ended up doing it with a table." - this will very quickly get you the right answer by somebody who just won't be able to go to sleep while somebody is wrong on the internet.

I wonder if LLMs can be taught outrage and spite.

Re: Social Skill Training with Large Language Models

#18

Post by maroonblazer »

kordlessagain wrote: post #11
jbandela1 wrote: post #5

One big concern I have with this is that the state of LLMs because of issues in where it can go horribly wrong, requires someone who is familiar with the subject to evaluate their output. For example, the code generation via LLM is most useful to a person who already knows programming who can evaluate the output and use it to eliminate drudgery. Previously, there was an article posted about using LLMs to generate pat…

Giving the underprivileged tools to work on themselves is a huge win for society, regardless of the fidelity of all interactions. There are a lot of bad therapists in the world, and not all of them are going to give good advice all the time. But, bad therapists can still provide some information that may be useful to the person learning. Why would we expect perfection from interacting with an LLM instead of a bad the…

> You have to get used to creating problems for yourself, then coping with the feelings that come, and pushing through to a solution.

A great description of any creative act, really.

Re: Social Skill Training with Large Language Models

#19

Post by Morelesshell »

jbandela1 wrote: post #5

One big concern I have with this is that the state of LLMs because of issues in where it can go horribly wrong, requires someone who is familiar with the subject to evaluate their output. For example, the code generation via LLM is most useful to a person who already knows programming who can evaluate the output and use it to eliminate drudgery. Previously, there was an article posted about using LLMs to generate pat…

The concern is real but you know what? The last company i was part of, had a TON wtf moments.

Like database unprotected on the internet (very short password too!), md5 for passwords, api key in the apps...

Re: Social Skill Training with Large Language Models

#20

Post by dazzawazza »

Use these models to enable people to be tolerant and value other communication styles... not to train everyone to communicate in identical fashions.

We seem to be working hard to homogenise humanity. Soon we'll all be trained by the same model to generate the same answers in the same way to the same beat. The era of the McHuman.

We seem to think we can use LLMs as an edge to make us richer or better than other humans. We're not competing with each other, we're competing with the LLM.

Post reply on HN