Have you ever had a *productive* conversation with someone you care about regarding the pitfalls of AI?
-
@ShaulaEvans I had this with my therapist when I refused consent for AI transcription. He was genuinely curious about my opinion because I work in tech. I pointed out there were security concerns because you don't know where the servers are located or what the laws arw there; I mentioned the possibility of hallucination and I didn't want that in my records; just general facts in a level way. Of course he legitimately wanted to learn, which possibly helped.
@commonst That's wonderful. What a service not just to the therapist but to all their other patients. Bravo.
-
Have you ever had a *productive* conversation with someone you care about regarding the pitfalls of AI? Not about big picture concerns like ethics, environmental impact, etc., but about the more immediate harms to the person, like inaccurate responses and the risk of AI psychosis.
I need to have a gentle, informative conversation in a non-shaming way with someone dear to me, and I'd love to learn from anyone who has done this successfully.
*AI Booster replies will be blocked.
@ShaulaEvans yes actually. It's rare but a few times.
It helps that I've had to use it professionally BUT am in QA so people occasionally listen to me when I warn about the dangers of trusting AI output without verification.
Some AI Bros didn't like to hear that which is probably why I'm not at that company anymore I think.
-
@commonst That's wonderful. What a service not just to the therapist but to all their other patients. Bravo.
@ShaulaEvans yeah, not sire if it changed his mind entirely, but it definitely made him think which I count as a win.
-
@ShaulaEvans I had this with my therapist when I refused consent for AI transcription. He was genuinely curious about my opinion because I work in tech. I pointed out there were security concerns because you don't know where the servers are located or what the laws arw there; I mentioned the possibility of hallucination and I didn't want that in my records; just general facts in a level way. Of course he legitimately wanted to learn, which possibly helped.
@commonst @ShaulaEvans if privacy is your concern, are you ok with self hosted models? it’s easy to download whisper and have it transcribe an mp3 without requiring remote servers
-
Have you ever had a *productive* conversation with someone you care about regarding the pitfalls of AI? Not about big picture concerns like ethics, environmental impact, etc., but about the more immediate harms to the person, like inaccurate responses and the risk of AI psychosis.
I need to have a gentle, informative conversation in a non-shaming way with someone dear to me, and I'd love to learn from anyone who has done this successfully.
*AI Booster replies will be blocked.
Stopped reading after the first seven words because they were clearly a personal attack. (He said, proving his own point!)
Good luck ...
-
Stopped reading after the first seven words because they were clearly a personal attack. (He said, proving his own point!)
Good luck ...
-
@commonst @ShaulaEvans if privacy is your concern, are you ok with self hosted models? it’s easy to download whisper and have it transcribe an mp3 without requiring remote servers
@eljojo @ShaulaEvans I have lots of concerns about AI, that is just one example. I would rather the only thing listening to my therapy sessions be my therapist.
-
Have you ever had a *productive* conversation with someone you care about regarding the pitfalls of AI? Not about big picture concerns like ethics, environmental impact, etc., but about the more immediate harms to the person, like inaccurate responses and the risk of AI psychosis.
I need to have a gentle, informative conversation in a non-shaming way with someone dear to me, and I'd love to learn from anyone who has done this successfully.
*AI Booster replies will be blocked.
@ShaulaEvans
I had limited success by pushing #aislop infatuated people to make a supervised session asking their lover questions in their very field of expertise. Agrotechnologist asked about the wheat treatments, hydrologist about the volumetric asessments of retention capacity and so on. Usually they initially said "Thats wonderful, it is almost correct". I then dig into the "almost" pushing one to do a dissection of the answer/summary and point where the "almost" is, and to correct all "almosts".Only then they start to grasp the levels of misinformation.
Final question usually is: will you allow "almost" trained ai robotic surgeon to open your chest? And why not.
-
Have you ever had a *productive* conversation with someone you care about regarding the pitfalls of AI? Not about big picture concerns like ethics, environmental impact, etc., but about the more immediate harms to the person, like inaccurate responses and the risk of AI psychosis.
I need to have a gentle, informative conversation in a non-shaming way with someone dear to me, and I'd love to learn from anyone who has done this successfully.
*AI Booster replies will be blocked.
@ShaulaEvans
I think listening to them first is crucial.If they feel heard and not judged, they'll be way more open to hear you as well.
Of course, this is extremely hard: The more we disagree, the less we want to hear their reasons.
But that opposition is the same for them, and the main obstacle to having productive discussions
I try to mentalize myself "This person has a good reason for thinking differently than me." and ask questions until I find what that is.
-
@eljojo @ShaulaEvans I have lots of concerns about AI, that is just one example. I would rather the only thing listening to my therapy sessions be my therapist.
-
@ShaulaEvans
I think listening to them first is crucial.If they feel heard and not judged, they'll be way more open to hear you as well.
Of course, this is extremely hard: The more we disagree, the less we want to hear their reasons.
But that opposition is the same for them, and the main obstacle to having productive discussions
I try to mentalize myself "This person has a good reason for thinking differently than me." and ask questions until I find what that is.
@DecksDark This is very grounded, thoughtful, caring advice. Thank you.
-
@ShaulaEvans
I had limited success by pushing #aislop infatuated people to make a supervised session asking their lover questions in their very field of expertise. Agrotechnologist asked about the wheat treatments, hydrologist about the volumetric asessments of retention capacity and so on. Usually they initially said "Thats wonderful, it is almost correct". I then dig into the "almost" pushing one to do a dissection of the answer/summary and point where the "almost" is, and to correct all "almosts".Only then they start to grasp the levels of misinformation.
Final question usually is: will you allow "almost" trained ai robotic surgeon to open your chest? And why not.
@ohir @ShaulaEvans That's a good idea. That said, what's especially frightening is that this is only something that'll probably only work for a short time yet. As you've described it, the exercise already requires a subject-matter expert with plenty of time.
But what happens when LLMs get a bit better, like completely accurate for hours of testing? Whatever people say about human error, AI chatbots still fail in completely different ways from people. LLMs are undergoing gradual improvement, but many expect immediate results. Like, I want to be enthusiastic about the prospect of AI improving, but we've moved seamlessly from them being to hoodwink almost nobody with their head screwed on, to being able to mislead almost all of the population, with only subject matter experts being able to tell. What happens in the next stage, when even the subject matter experts are no longer able to tell AI apart from their peers in the field, but the LLMs continue to produce subtly dangerous output?
I have no answers, just fear...
-
@ohir @ShaulaEvans That's a good idea. That said, what's especially frightening is that this is only something that'll probably only work for a short time yet. As you've described it, the exercise already requires a subject-matter expert with plenty of time.
But what happens when LLMs get a bit better, like completely accurate for hours of testing? Whatever people say about human error, AI chatbots still fail in completely different ways from people. LLMs are undergoing gradual improvement, but many expect immediate results. Like, I want to be enthusiastic about the prospect of AI improving, but we've moved seamlessly from them being to hoodwink almost nobody with their head screwed on, to being able to mislead almost all of the population, with only subject matter experts being able to tell. What happens in the next stage, when even the subject matter experts are no longer able to tell AI apart from their peers in the field, but the LLMs continue to produce subtly dangerous output?
I have no answers, just fear...
@ohir @ShaulaEvans oh, and to respond to this:
will you allow "almost" trained ai robotic surgeon to open your chest?
Depends on how desperate for the operation I am. Destroying entire professional specialisms economically in order to later replace them with substandard slop machines is already a page in the AI-peddlers playbook. Lots of 2D digital artists I follow have already basically said they're suffering the results of that process at this very moment.
-
@eljojo @ShaulaEvans I have lots of concerns about AI, that is just one example. I would rather the only thing listening to my therapy sessions be my therapist.
@commonst @ShaulaEvans are these concerns something that you’re interested in addressing? or just good excuses to reject AI?
what are your other concerns? maybe, just like “where the servers are”, something can be done about them.
what do you mean when you say AI “listens” to conversations, is that different than your smartphone “listening” for you to say “hey google”? do you turn your phone off during therapy?
-
R relay@relay.an.exchange shared this topic
