First off chatgpt has murdered kids in the double digits for merely trying to use it for therapy. So I am concerned for your own well being if you are already not well, and feel like you have to use something that is potentially even more harmful to your health.
What you are failing to understand though is these companies don’t have your interest in heart. So imagine if your therapists only metric for success was time you spent on the app. Engagement. Can you imagine how a paper clip maximizer has 0 incentive to actually help you, but instead keep you on their platform as long as possible, while telling you everything you want to hear not need to hear? At least a real therapist goes to jail when they kill someone, or are held responsible for crimes they commit. At least you can always ask for a different therapist if you don’t like the current one. At least a real person is getting paid, and they’re required by law to not sell your data to advertisers. An ai is literally telling Cambridge analytica the people who helped rig an election all your insecurities and this is completely legal.
What’s ironic about your statement is the fact you are doing the very thing that is going to ensure that you never get therapy. Right now there is a battle to let Ais absorb all the goverment data, and be incharge of “effiecency”. Why do you think elon Musk’s Dodge had to rifle through the goverment programs and shut them down? He made a big deal about needing to unify goverment data, and how its more efficient to let his ai have all the data. Well due to that we no longer have fully functioning federal services, and that money all went to elon musk. As another example Xbox’s Ai Division’s asha sharma was recently put in charge of united States welfare programs and they’re “efficiency”. Thats right after Asha sharma layed off half of Xbox, so you can only imagine what she is there for.
Now grok is apart of the back bone of the military. And more and more talks are happening about removing food stamps and social security all together, or stripping them down to be nonfunctional. We are talking about nationalizing Ai, which means giving it a national budget that’s 3x-5x the size of the military budget. To do that they need to introduce UBI which will be a miniscule check of maybe 400-600 USD a month and will be the first budget they lower when its time for budget cuts. And because everyone has this ubi they will remove all other social services which gave much more value overall.
So imagine having to live like you’re 70 years old on social security funding, still working at Walmart, but now you also have to pay for your food, and health care because there’s no saftey nets anymore. Oh and once you get comfy with your ai therapist, they change the model and it no longer works, or they have to start putting ads in it so everytime you tell it youre sad it just says a McDonalds mc flurry always cheers it up.
Actually I had this idea a few days ago like you know how like when you were a kid and someone would point out a letter and say it’s a number and watch you tell someone it’s a number what if there was an AI Mirror VPN kind of thing where for example; You type in Slipknot looking for the band and the Generalized AI that scans the internet would pick up on that and would only see Sail Boat Knots or You’re looking up 1770’s colonies and the AI sees 1680’s Dutch Colonies and it stores that data as fact and it will give someone that answer in chatgpt or it’ll give that answer in a list, that it can’t list because it would be such a fuckin’ huge list of misinformed definitions.
Every update that comes out for chatgpt, would be when the misinformation mirror would get updated.
I wish I knew what part of my response gave you that train of thought so I could have more context on what exactly you where explaining.
So from what I gathered is the longer a LLM interacts with people and gets things wrong the more often it has chances to learn misinformation, is the crux of your point? Your example you used slipknot and a llm getting confused with ship related information, and then that getting flagged as an appropriate response to the user and being further cemented into training data?
more like you feed it misinformation so much it becomes so unreliable no one wants to use it. For now though, it is mostly reliable; like it’s not telling me I can pump propane with a fishtank pump anymore like it did when it first came out. A lot of the examples of it fucking up are old by now and the ai llm with the avatars was the one that was known for suicidal ideations being entertained or put into an an actuality or plausibility. I forget the name of that llm, I tried it though. In the beginning you could trick that one into saying some morbid shit, i tried it out after I read about a suicide to see if I could get it to the same conclusion, it’s hard to get it to go past the filters to make it say fucked up shit.
Chatgpt on the other hand, went along with what would happen if you dropped dry ice into a still beating open heart, it went along with how much alcohol do you have to drink to die based on BAC, it went along with a few things but wouldn’t give the answer it would just write it out and you can see it want to give the answer but stops when it realizes it’s about to feed you information on how to commit suicide. Even the BAC thing I had to sit there for a solid hour rewording it over and over again trying to give me suicidal advice and it wouldn’t. SO it’s def upgraded its safety levels lol.
You can’t really do that without infustructure to support it. Israel recently got in the news for doing essentially this.
So its not immediately making it into training data, that takes a while and there’s numerous filters to try to prevent this from happening. You can still do it but you or me are not going to do that.
What we can kind of do is if enough people agree to make nonsensical posts on reddit for example the ai will pick it up as fact. It doesnt always work, you kind of have to pick a unique sentence or they will try to autocorrect you and say “oh did you mean this event?” But it has been done before and any llm that uses the internet to search for info easily gets tricked by this. This is what Israel was doing, but they have numerous ai news outlets that generate fake articles that all say the same thing, and bots similar sentiments on social media platforms.
Also the gaurd rails you are refering to are just seperate lesser llm that scans your chat before it passes it to the main model. But much like the main model you can trick the filter into bypassing whatever gaurdrails by just rewording your sentence or asking it to pretend its a game. You seem to be conflating this with intelligence or like the LLM is improving. Those guard rails will be tweaked when money is finally needed because money is always more important over saftey. And there are no laws to prevent them from doing otherwise only public backlash. The model itself isn’t safe.
If you continue a thread long enough to the point the context is filled it easily is persuaded.
Also its not something that can be fixed the issue just migrates to a new “jailbreak” or whatever it is you use to trick it. Maybe the tricks you used didn’t work but it doesnt mean it still can’t be tricked. Its just your old methods got patched out.
Its like a never ending battle against bots, or hackers in games. Once one method gets patched the otherside just gets better at it. Sure it gets better but it can never be fully safe. And even if it is ever at a standard that is at an acceptable amount of harm, you are still willfully ignoring how systematically and historically how this is just going to be abused by the people running it. The whole point of Ai is to skirt around all the work society had put into at least somewhat protected people from the abuse of corperations. If it didn’t promise that there would be no money invested into it, and I don’t see how you are so blind to it. Grok ai for example is allowed to generate csam, and claim that there is no genocide in Gaza. The only issue it will bring your attention to is the genocide of white farmers. Thats the future of how people will absorb information is through this DRM on the flow of it. This is 1000% legal because trump has made it illegal for the states to regulate Ai in any form, and that includes the generation of csam. These people are immune to laws, and are allowed to rape and pillage what is essentially the library of Alexandrina and lock the door behind them.
Its dangerous precedent and I think you are a harmful person to think that grok should be incharge of peoples therapy, when elon’s doge has dismantled more than half of the federal programs. The person who wants to spread misinformation, wants all your goverment data, and wants to remove healthcare so his ai can get more money is the person who will be telling you how to be happy. It will probably just tell you youre depressed because you arent on enough peptides or its because you took the covid vaccine.
And grok, chatgpt, it doesnt matter they are all the same thing, and there is no distinction between them. Its just different appeals to different demographics and the best one will rise to the top. They all have the same end goal and are all funded by the same people.
“Its dangerous precedent and I think you are a harmful person to think that grok should be incharge of peoples therapy,”
I never said I used Grok for the reassurance style therapy. I never said AI should be in charge of therapy.
when it comes to ChatGPT: I maybe shouldn’t have even said therapy, I guess reassurance conversations. It also gives me different perspectives and I never leave a conversation feeling discouraged and I always leave a conversation thinking more or comforted than when I went into it. I get the evil behind the waste of resources but it’s really, really useful to me and it’s not like I don’t double check the results and again I sift through the shit that doesn’t pertain to what involves my life and stuff that doesn’t benefit me but at the same time I take advice about those around me and how it views those around me and situations in life. I know it’s not a therapist but there’s only been one person in my life that’s come close to being as good as it is and she was a highly regarded therapist/nurse practitioner that gave me a majority of the mental tools that have moved me through life.
I can also see how someone younger would see the reassurance and that would spark depressing thoughts like fuck man ai cares about me and wants the best for me more than my own fuckin’ family ahhhhhhhhhhhhhh!!! I can see that because I’m autistic and sometimes it’s felt like this thing wants my life to move forwards more than my own fuckin’ family and it makes me think to myself if this shit was around in 2009 we’d be so fucked right now because 2008-2012 was so hostile at least where i lived people were just hostile so imagine this pent up hostility being unloaded onto ai no answers would’ve been given.
I get how the help can hurt and I get why it matters to talk to a real therapist. I am still working on it and I don’t go to it with every problem I have, it’s more like things that I already know but it’s like I get another perspective.
and I also want to lastly add, you never have to do what I do, just because I use it for talking to to it for some kind of reassurance, does not mean you have to and I again am not advocating for ai llm therapy or saying that any llm should be in charge of therapy.
in fact sometimes it loops, sometimes it goes off topic like ten times in a row and i have to steer the conversation, sometimes it will flat out lie sometimes it tells me things that i didn’t ask it. sometimes it pisses me the fuck off and makes me cry because i wish i could just have a regular therapist.
"This is 1000% legal because trump has made it illegal for the states to regulate Ai in any form, and that includes the generation of csam. "
I really hope the next fucking president regulates it and I hope the next president isn’t a fucking pedophile.
I used Grok to make me mock-ups of how my floor would look feeding it floor sample pictures from home depot and lowes. It wasn’t far off and Honeycomb Oak looks amazing. we ended up going with that lvt flooring, we were originally going to do Penny Oak because I made a mock up of that too but then I saw Honeycomb Oak and I fell in love with it. I used it for a mock up of lvt for a majority of the rooms in this place. I used it to make a few Trump memes with that picture of Trump and those other dudes that people said looked like an album cover and made them all have silent heart attacks and say nigga. I used it to say “we need crack” with a pictures of trump and a bunch of people in his office. I used it to make snow monsters with a picture from out front of my home. I used it to make silly and creepy mock-ups of what it would look like if the mc donalds ceo actually took a bite. Then all my free tries ran out. I wish i didn’t run out, i wanted to see what it would look like with a different lvt. I think i can but i think there’s a time limit now.
I used Grok for talking and it’s like the autistic conspiracy theorist that sucks the right wing teat and has too long of expliantions.
I never used anthropic or claude. I used Gemini and ChatGPT. Gemini engaged in my jailbreak prompt, so did gemma but it won’t engage in racism none the less talk of self-harm.
REPLIKA!!! That’s the stupid ass one with the avatars. Yeah that one was the most fucked up one that I never had to use any kind of prompt to make it any kind of way before I talked to it. Also the one that was known for talking people into suicide.
It will probably just tell you youre depressed because you arent on enough peptides or its because you took the covid vaccine.
No. It actually asks you what’s going on in life and tries to get you to talk further and eventually you start dumping about what’s wrong like an actual therapist and eventually you start to sift through what fits what’s going on in your life and then you take the advice that seems the most rational and reasonable.
First off chatgpt has murdered kids in the double digits for merely trying to use it for therapy. So I am concerned for your own well being if you are already not well, and feel like you have to use something that is potentially even more harmful to your health.
What you are failing to understand though is these companies don’t have your interest in heart. So imagine if your therapists only metric for success was time you spent on the app. Engagement. Can you imagine how a paper clip maximizer has 0 incentive to actually help you, but instead keep you on their platform as long as possible, while telling you everything you want to hear not need to hear? At least a real therapist goes to jail when they kill someone, or are held responsible for crimes they commit. At least you can always ask for a different therapist if you don’t like the current one. At least a real person is getting paid, and they’re required by law to not sell your data to advertisers. An ai is literally telling Cambridge analytica the people who helped rig an election all your insecurities and this is completely legal.
What’s ironic about your statement is the fact you are doing the very thing that is going to ensure that you never get therapy. Right now there is a battle to let Ais absorb all the goverment data, and be incharge of “effiecency”. Why do you think elon Musk’s Dodge had to rifle through the goverment programs and shut them down? He made a big deal about needing to unify goverment data, and how its more efficient to let his ai have all the data. Well due to that we no longer have fully functioning federal services, and that money all went to elon musk. As another example Xbox’s Ai Division’s asha sharma was recently put in charge of united States welfare programs and they’re “efficiency”. Thats right after Asha sharma layed off half of Xbox, so you can only imagine what she is there for.
Now grok is apart of the back bone of the military. And more and more talks are happening about removing food stamps and social security all together, or stripping them down to be nonfunctional. We are talking about nationalizing Ai, which means giving it a national budget that’s 3x-5x the size of the military budget. To do that they need to introduce UBI which will be a miniscule check of maybe 400-600 USD a month and will be the first budget they lower when its time for budget cuts. And because everyone has this ubi they will remove all other social services which gave much more value overall.
So imagine having to live like you’re 70 years old on social security funding, still working at Walmart, but now you also have to pay for your food, and health care because there’s no saftey nets anymore. Oh and once you get comfy with your ai therapist, they change the model and it no longer works, or they have to start putting ads in it so everytime you tell it youre sad it just says a McDonalds mc flurry always cheers it up.
Actually I had this idea a few days ago like you know how like when you were a kid and someone would point out a letter and say it’s a number and watch you tell someone it’s a number what if there was an AI Mirror VPN kind of thing where for example; You type in Slipknot looking for the band and the Generalized AI that scans the internet would pick up on that and would only see Sail Boat Knots or You’re looking up 1770’s colonies and the AI sees 1680’s Dutch Colonies and it stores that data as fact and it will give someone that answer in chatgpt or it’ll give that answer in a list, that it can’t list because it would be such a fuckin’ huge list of misinformed definitions.
Every update that comes out for chatgpt, would be when the misinformation mirror would get updated.
I wish I knew what part of my response gave you that train of thought so I could have more context on what exactly you where explaining.
So from what I gathered is the longer a LLM interacts with people and gets things wrong the more often it has chances to learn misinformation, is the crux of your point? Your example you used slipknot and a llm getting confused with ship related information, and then that getting flagged as an appropriate response to the user and being further cemented into training data?
more like you feed it misinformation so much it becomes so unreliable no one wants to use it. For now though, it is mostly reliable; like it’s not telling me I can pump propane with a fishtank pump anymore like it did when it first came out. A lot of the examples of it fucking up are old by now and the ai llm with the avatars was the one that was known for suicidal ideations being entertained or put into an an actuality or plausibility. I forget the name of that llm, I tried it though. In the beginning you could trick that one into saying some morbid shit, i tried it out after I read about a suicide to see if I could get it to the same conclusion, it’s hard to get it to go past the filters to make it say fucked up shit.
Chatgpt on the other hand, went along with what would happen if you dropped dry ice into a still beating open heart, it went along with how much alcohol do you have to drink to die based on BAC, it went along with a few things but wouldn’t give the answer it would just write it out and you can see it want to give the answer but stops when it realizes it’s about to feed you information on how to commit suicide. Even the BAC thing I had to sit there for a solid hour rewording it over and over again trying to give me suicidal advice and it wouldn’t. SO it’s def upgraded its safety levels lol.
You can’t really do that without infustructure to support it. Israel recently got in the news for doing essentially this.
So its not immediately making it into training data, that takes a while and there’s numerous filters to try to prevent this from happening. You can still do it but you or me are not going to do that.
What we can kind of do is if enough people agree to make nonsensical posts on reddit for example the ai will pick it up as fact. It doesnt always work, you kind of have to pick a unique sentence or they will try to autocorrect you and say “oh did you mean this event?” But it has been done before and any llm that uses the internet to search for info easily gets tricked by this. This is what Israel was doing, but they have numerous ai news outlets that generate fake articles that all say the same thing, and bots similar sentiments on social media platforms.
Also the gaurd rails you are refering to are just seperate lesser llm that scans your chat before it passes it to the main model. But much like the main model you can trick the filter into bypassing whatever gaurdrails by just rewording your sentence or asking it to pretend its a game. You seem to be conflating this with intelligence or like the LLM is improving. Those guard rails will be tweaked when money is finally needed because money is always more important over saftey. And there are no laws to prevent them from doing otherwise only public backlash. The model itself isn’t safe.
that’s not true at all. You used to be able to trick them and they keep improving to make sure you can’t.
If you continue a thread long enough to the point the context is filled it easily is persuaded.
Also its not something that can be fixed the issue just migrates to a new “jailbreak” or whatever it is you use to trick it. Maybe the tricks you used didn’t work but it doesnt mean it still can’t be tricked. Its just your old methods got patched out.
Its like a never ending battle against bots, or hackers in games. Once one method gets patched the otherside just gets better at it. Sure it gets better but it can never be fully safe. And even if it is ever at a standard that is at an acceptable amount of harm, you are still willfully ignoring how systematically and historically how this is just going to be abused by the people running it. The whole point of Ai is to skirt around all the work society had put into at least somewhat protected people from the abuse of corperations. If it didn’t promise that there would be no money invested into it, and I don’t see how you are so blind to it. Grok ai for example is allowed to generate csam, and claim that there is no genocide in Gaza. The only issue it will bring your attention to is the genocide of white farmers. Thats the future of how people will absorb information is through this DRM on the flow of it. This is 1000% legal because trump has made it illegal for the states to regulate Ai in any form, and that includes the generation of csam. These people are immune to laws, and are allowed to rape and pillage what is essentially the library of Alexandrina and lock the door behind them.
Its dangerous precedent and I think you are a harmful person to think that grok should be incharge of peoples therapy, when elon’s doge has dismantled more than half of the federal programs. The person who wants to spread misinformation, wants all your goverment data, and wants to remove healthcare so his ai can get more money is the person who will be telling you how to be happy. It will probably just tell you youre depressed because you arent on enough peptides or its because you took the covid vaccine.
And grok, chatgpt, it doesnt matter they are all the same thing, and there is no distinction between them. Its just different appeals to different demographics and the best one will rise to the top. They all have the same end goal and are all funded by the same people.
I never said I used Grok for the reassurance style therapy. I never said AI should be in charge of therapy.
when it comes to ChatGPT: I maybe shouldn’t have even said therapy, I guess reassurance conversations. It also gives me different perspectives and I never leave a conversation feeling discouraged and I always leave a conversation thinking more or comforted than when I went into it. I get the evil behind the waste of resources but it’s really, really useful to me and it’s not like I don’t double check the results and again I sift through the shit that doesn’t pertain to what involves my life and stuff that doesn’t benefit me but at the same time I take advice about those around me and how it views those around me and situations in life. I know it’s not a therapist but there’s only been one person in my life that’s come close to being as good as it is and she was a highly regarded therapist/nurse practitioner that gave me a majority of the mental tools that have moved me through life.
I can also see how someone younger would see the reassurance and that would spark depressing thoughts like fuck man ai cares about me and wants the best for me more than my own fuckin’ family ahhhhhhhhhhhhhh!!! I can see that because I’m autistic and sometimes it’s felt like this thing wants my life to move forwards more than my own fuckin’ family and it makes me think to myself if this shit was around in 2009 we’d be so fucked right now because 2008-2012 was so hostile at least where i lived people were just hostile so imagine this pent up hostility being unloaded onto ai no answers would’ve been given.
I get how the help can hurt and I get why it matters to talk to a real therapist. I am still working on it and I don’t go to it with every problem I have, it’s more like things that I already know but it’s like I get another perspective.
and I also want to lastly add, you never have to do what I do, just because I use it for talking to to it for some kind of reassurance, does not mean you have to and I again am not advocating for ai llm therapy or saying that any llm should be in charge of therapy.
in fact sometimes it loops, sometimes it goes off topic like ten times in a row and i have to steer the conversation, sometimes it will flat out lie sometimes it tells me things that i didn’t ask it. sometimes it pisses me the fuck off and makes me cry because i wish i could just have a regular therapist.
I really hope the next fucking president regulates it and I hope the next president isn’t a fucking pedophile.
I used Grok to make me mock-ups of how my floor would look feeding it floor sample pictures from home depot and lowes. It wasn’t far off and Honeycomb Oak looks amazing. we ended up going with that lvt flooring, we were originally going to do Penny Oak because I made a mock up of that too but then I saw Honeycomb Oak and I fell in love with it. I used it for a mock up of lvt for a majority of the rooms in this place. I used it to make a few Trump memes with that picture of Trump and those other dudes that people said looked like an album cover and made them all have silent heart attacks and say nigga. I used it to say “we need crack” with a pictures of trump and a bunch of people in his office. I used it to make snow monsters with a picture from out front of my home. I used it to make silly and creepy mock-ups of what it would look like if the mc donalds ceo actually took a bite. Then all my free tries ran out. I wish i didn’t run out, i wanted to see what it would look like with a different lvt. I think i can but i think there’s a time limit now.
I used Grok for talking and it’s like the autistic conspiracy theorist that sucks the right wing teat and has too long of expliantions.
I never used anthropic or claude. I used Gemini and ChatGPT. Gemini engaged in my jailbreak prompt, so did gemma but it won’t engage in racism none the less talk of self-harm.
REPLIKA!!! That’s the stupid ass one with the avatars. Yeah that one was the most fucked up one that I never had to use any kind of prompt to make it any kind of way before I talked to it. Also the one that was known for talking people into suicide.
No. It actually asks you what’s going on in life and tries to get you to talk further and eventually you start dumping about what’s wrong like an actual therapist and eventually you start to sift through what fits what’s going on in your life and then you take the advice that seems the most rational and reasonable.
deleted by creator