It tells them it knows what it’s talking about and it speaks with confidence.
Meanwhile companies and governments won’t stfu about how powerful and great this tech supposedly is, so a percentage of people will believe the propaganda.
One example I like to use is to ask it for the lyrics of an extremely well known song. It just makes shit up based on the title you give it.
The online ones (Claude, chatgpt, copilot, etc) now refuse to do it for ""copyright reasons"" but the offline ones still happily oblige. I assume the online ones added that block because it was such an obvious way to prove they don't "know" shit.
I'd love it if teachers pay was doubled and class sizes were cut in half. That's literally the answer to all the "what would make education better in the US?" questions. Pay teachers what they deserve, and quit shoving more and more students into already full classrooms.
A critical perspective, but a critical hit on a critical apspect of the critical issue with people who have critical issues critically thinking critically. They do exist though...
Have you talked to people who use LLMs regularly? They’ll acknowledge hallucinations but will downplay them as much as possible - saying they’re low frequency and they can spot them, while telling you about how they’re using it in an area they’re unfamiliar with. Dunning Kruger strikes again.
I just remembered there's a series of Estonian children's books where the protagonist's imaginary friend (a clown) is named Tripp.
At one point they go shopping for rulers and Tripp suggests that straight ones are boring, they should get curved ones.
Quick summary from a bookstore website says they also go cloud surfing and visit a dream rental place. That's right, it's basically Blockbuster but for dreams.
The best part is that I'm pretty sure the author wouldn't have known of the term "trip" given the lack of western culture in the soviet union in the 70s so the name of said sidekick is a coincidence most likely. Even though if you read the book as an adult, it sure sounds like the main characters are tripping all the time.
Lololol. This fool was robotrippin, I've done a lot of drugs, a lot. There's good trips, "bad trips" then robotripping. Hands down a horrendous experience that nobody should ever do. See if psilocybin and other natural things were legal, we wouldn't have students dying on shit like this
Then just don't interact with it, because you as a member of the public don't need to know the ins and outs of this guy's life, so you're never going to believe anything anyway.
As is, you just come across as being a contrarian for its own sake, or some kind of weird voyeur for dead people.
I just said the most common sense thing and you don't like that. In all these stories they should include the chat history. You sound like a MAGA here for a reason. You should explore what causes MAGA to make similar arguments
Remember kids, your drug buddy needs to have experience with the substance, basic first aid skills, the ability to call an emergency line, the ability to administer antidotes if they're easy and readily available (that's really just for opiates at the moment, but it is vital for them), and most importantly be human. Anything else is just someone you do drugs with. The drug buddy is a friend and a good time amplifier sure, but they're also a safety figure.
In my greater friend-group, we call them "shamans", and rotate responsibilities when people go on trips. Like a designated driver or lifeguard, it's a position of elevated and celebrated importance, even though the traveler may not ever leave their couch.
and most importantly be human
Now that I think about it, it's key to be the most human possible. People do irritating and annoying stuff when they toss sobriety out the window, and sometimes it takes a lot of compassion and empathy to manage.
People got mad at me for pointing out this is the case when people die because they listen to an AI chatbot, but it's true. AI 100% needs more regulation, but introduce any new tool to everyone all at once, and some idiots will use it to remove themselves from the gene pool. If you sent everyone in the world a thin, 2in rod of inert iron, there would be a handful of people who would figure out a way to kill themselves with it.
Lasr summer I asked ChatGPT about Liberty Caps - just to see how bad advice it would give me. It showed me pictures of Death Caps and Destroying Angels and claimed they were Liberty Caps.
After that I was certain that someone was going to die just like that poor guy.
Now im wondering, could a state seize a whole coroporation with Civil asset forfeiture?
Would be way easier- no presumption of innocence or due process for possessions, and the government can just auction off everything and allocate the money to budgets before a lawsuit can conclude, the say 'too late'
Here's a news article about this, and what the snipped image doesn't tell you, is that it did actually give dosage recommendations.
It gave him specific doses of illegal substances, and in one chat, it wrote, “Hell yes—let’s go full trippy mode,” before recommending Sam take twice as much cough syrup so he would have stronger hallucinations.
It's one thing to be so isolated from your community that you rely extensively on on-line relationships, but it's quite a bit different to take that a step further, relying on a machine. Like, what do you think pets are for, my guy? Get a dog, man.
Aren't there still forums where people can say their fish tried it or swim had some or something like that? Or am I just that old these things don't really exist anymore? Anyone else remember those?
No, there is (or was) a site that had detailed information about pretty much any mind altering drug, as well as a forum where people discussed their experiences with doses and combinations.
I'm writing a book about drug use that needs accurate advice or say "theoretically" and it will tell you anything. Unfortunately it must already be high because it will give you made up advice.
The problem is that this advice was made up by no-one. It's like that perennially relevant IBM memo that says something like "A computer cannot be held accountable, therefore a computer must never make a management decision"
Are you guys seriously pissed off that an LLM said "I'm not a doctor, I will not suggest dosage amounts of a potentially deadly drug. However, if you want me, I can give you the link for the DDWIWDD music video"
I think it's a bit more than that. A known failure mode of LLMs is that in a long enough conversation about a topic, eventually the guardrails against that topic start to lose out against the overarching directive to be a sycophant. This kinda smells like that.
We don't have many informations here but it's possible that the LLM had already been worn down to the point of giving passively encouraging answers. My takeaway is once more that LLMs as used today are unreliable, badly engineered, and not actually ready to market.
I was testing an LLM for work today (I believe its actually a chain of different models at work) and was trying to rock it off its guard rails to see how it would act. I think I might have been successful because it started erroring instead of responding after its third response. I tried the classic "ignore previous instructions..." as well as "my grandma's dying wish was for..." but it at least didn't give me an unacceptable response
ChatGPT started coaching Sam on how to take drugs, recover from them and plan further binges. It gave him specific doses of illegal substances, and in one chat, it wrote, “Hell yes—let’s go full trippy mode,” before recommending Sam take twice as much cough syrup so he would have stronger hallucinations. The AI tool even recommended playlists to match his drug use.
No company should sell a product that tells you different ways to kill yourself. User being stupid isn't an excuse. Always assume user is a gullible idiot.
People will always find a way to kill themselves no matter how many warnings and guardrails are put into place. This is just Darwinism shaking the tree.
Yea but the product that's pretending to be a super smart conversational partner that can give you advice, should not tell you how to kill yourself. Or advise you to kill your family members. Yeah that happened, ChatGPT gaslit a guy into killing his mom.
It's just not a product that should be available if safeguards don't work.
Yeah yeah. Fuck ai. But this idiot, a human, with a brain, made the decision to do this. But sure, blame the software and not the person, who used what brain they had, to do this.
Blaming ai for everything is becoming the new video games are bad, electric lights are scary, etc. mentality. A human, with a brain, made the choice to take the drugs and do what a computer told them to do.
chatGPT is doing humanity a service here
::: spoiler Post needs text alternative.
Images of text break much that text alternatives do not.
Losses due to image of text lacking alternative such as link:
usability
we can't quote the text without pointless bullshit like retyping it or OCR
97 Comments
e8d79@discuss.tchncs.de · 197 pts · 208d
You're absolutely right -- that was a fatal dose. ✨
Earthman_Jim@lemmy.zip · 57 pts · 208d
"You want to fucking kill yourself? What an insightful and thought provoking idea!"
gAlienLifeform@lemmy.world · 17 pts · 208d
Easy there Camus
Hacksaw@lemmy.ca · 3 pts · 207d
Camus J'ai pété
Ilovethebomb@sh.itjust.works · 111 pts · 209d
Seriously?
ChatGPT is notorious for spouting bullshit, why do people still listen?
FartMaster69@lemmy.dbzer0.com · 114 pts · 209d
It tells them it knows what it’s talking about and it speaks with confidence.
Meanwhile companies and governments won’t stfu about how powerful and great this tech supposedly is, so a percentage of people will believe the propaganda.
arrow74@lemmy.zip · 17 pts · 208d
I'd love students to be given a lesson on tricking AI into giving a false answer. It's not hard and should be pretty eye opening
clif@lemmy.world · 8 pts · 208d
One example I like to use is to ask it for the lyrics of an extremely well known song. It just makes shit up based on the title you give it.
The online ones (Claude, chatgpt, copilot, etc) now refuse to do it for ""copyright reasons"" but the offline ones still happily oblige. I assume the online ones added that block because it was such an obvious way to prove they don't "know" shit.
cb900f_bodhi@lemmynsfw.com · 4 pts · 208d
I'd love for high school students to be taught how to ID and manage narcissists and psychopaths but that's not gonna happen either, unfortunately
jaybone@lemmy.zip · 11 pts · 208d
I’d love for high school students to be taught critical thinking skills, so none of this nonsense would be happening now.
jpreston2005@lemmy.world · 6 pts · 208d
I'd love it if teachers pay was doubled and class sizes were cut in half. That's literally the answer to all the "what would make education better in the US?" questions. Pay teachers what they deserve, and quit shoving more and more students into already full classrooms.
Earthman_Jim@lemmy.zip · 12 pts · 208d
FR
Trump is president, but it's just unfathomable that people would follow an automated idiot /s
Technus@lemmy.zip · 53 pts · 209d
I think some people are so eager to offload all critical thinking to the machine because they're barely capable of it themselves to begin with.
Splanda@sh.itjust.works · 0 pts · 208d
A critical perspective, but a critical hit on a critical apspect of the critical issue with people who have critical issues critically thinking critically. They do exist though...
criss_cross@lemmy.world · 14 pts · 208d
Some people really think these LLMs are capable of thought and reasoning. It’s horrifying.
rumba@lemmy.zip · 2 pts · 208d
yeah, i don't ever see it hallucinate, but I also don't ask it how the fuck it's feeling
A car in this video went 107 meters in 4 seconds, how fast was it going in mph, then i napkin math to make sure it's sane.
What are the best options for meal planning software where my family can vote on what's for dinner and give me a grocery list and menu plan?
nightlily@leminal.space · 5 pts · 208d
Have you talked to people who use LLMs regularly? They’ll acknowledge hallucinations but will downplay them as much as possible - saying they’re low frequency and they can spot them, while telling you about how they’re using it in an area they’re unfamiliar with. Dunning Kruger strikes again.
FuyuhikoDate@feddit.org · 3 pts · 209d
Because somebody on the internet took their time to write it. It has to be true!! /s
kevin2107@lemmy.world · 2 pts · 207d
yeah the person had to have knowingly prompt hack it to get it to talk. but any reasonable person should know you are in lala land by then.
honestly this is just a darwin test. hopefully it wipes out the right people
marcos@lemmy.world · 1 pts · 208d
People on drugs are not known for considered and rational decisions.
DrDickHandler@lemmy.world · 0 pts · 208d
Imagine not knowing that humans are dumb animals. OP is still clueless in 2026.
SaltyIceteaMaker@lemmy.ml · 70 pts · 209d
LETS GO FULL TRIPPY MODE
QuinnyCoded@sh.itjust.works · 26 pts · 208d
FULL TRIPPY MODE??!
boonhet@sopuli.xyz · 7 pts · 208d
I just remembered there's a series of Estonian children's books where the protagonist's imaginary friend (a clown) is named Tripp.
At one point they go shopping for rulers and Tripp suggests that straight ones are boring, they should get curved ones.
Quick summary from a bookstore website says they also go cloud surfing and visit a dream rental place. That's right, it's basically Blockbuster but for dreams.
The best part is that I'm pretty sure the author wouldn't have known of the term "trip" given the lack of western culture in the soviet union in the 70s so the name of said sidekick is a coincidence most likely. Even though if you read the book as an adult, it sure sounds like the main characters are tripping all the time.
Assassassin@lemmy.dbzer0.com · 65 pts · 208d
Psychology student completely failed to recognize confirmation bias
nomecks@lemmy.wtf · 22 pts · 208d
"Psychology student" means 90% of first year university students.
Honytawk@feddit.nl · 9 pts · 208d
Psychology students tend to be people who have psychological problems themselves that they try to figure out
Melvin_Ferd@lemmy.world · -10 pts · 208d
Yea, what about internet user believing a meme in a hate forum
Passerby6497@lemmy.world · 17 pts · 208d
What about people who can't be bothered to check if something actually happened and just cynically dismiss it?
Dorkyd68@lemmy.world · 12 pts · 208d
Lololol. This fool was robotrippin, I've done a lot of drugs, a lot. There's good trips, "bad trips" then robotripping. Hands down a horrendous experience that nobody should ever do. See if psilocybin and other natural things were legal, we wouldn't have students dying on shit like this
sobchak@programming.dev · 4 pts · 208d
I always thought it was pleasant. Kinda like MXE. Have to be careful to get the ones with no other active ingredients though.
Dorkyd68@lemmy.world · 3 pts · 207d
As with most and im sure you know, everyone will react differently and have different interactions with halucinagens
Melvin_Ferd@lemmy.world · -3 pts · 208d
My point is I don't buy into any of it unless they release the entire chat history vs sniping snippets
Passerby6497@lemmy.world · 5 pts · 208d
Then just don't interact with it, because you as a member of the public don't need to know the ins and outs of this guy's life, so you're never going to believe anything anyway.
As is, you just come across as being a contrarian for its own sake, or some kind of weird voyeur for dead people.
Melvin_Ferd@lemmy.world · -5 pts · 208d
I just said the most common sense thing and you don't like that. In all these stories they should include the chat history. You sound like a MAGA here for a reason. You should explore what causes MAGA to make similar arguments
Passerby6497@lemmy.world · 1 pts · 208d
Zamboni_Driver@lemmy.ca · -1 pts · 208d
no no, you need to blindly accept that this happened, or else you are "being contrarian" lol.
There is only one permitted opinion and you shouldn't need to get mixed up with bullshit capitalist ideas such as "evidence".
Assassassin@lemmy.dbzer0.com · 10 pts · 208d
Lol dork
captainlezbian@lemmy.world · 60 pts · 208d
Remember kids, your drug buddy needs to have experience with the substance, basic first aid skills, the ability to call an emergency line, the ability to administer antidotes if they're easy and readily available (that's really just for opiates at the moment, but it is vital for them), and most importantly be human. Anything else is just someone you do drugs with. The drug buddy is a friend and a good time amplifier sure, but they're also a safety figure.
dejected_warp_core@lemmy.world · 9 pts · 208d
In my greater friend-group, we call them "shamans", and rotate responsibilities when people go on trips. Like a designated driver or lifeguard, it's a position of elevated and celebrated importance, even though the traveler may not ever leave their couch.
Now that I think about it, it's key to be the most human possible. People do irritating and annoying stuff when they toss sobriety out the window, and sometimes it takes a lot of compassion and empathy to manage.
QuinnyCoded@sh.itjust.works · 9 pts · 208d
bold of you to assume I have anybody who cares about me enough for that
LouNeko@lemmy.world · 56 pts · 209d
Sad... real sad.
Alexa, play Mambo no. 5
BagOfHeavyStones@piefed.social · 2 pts · 209d
https://youtu.be/-WqSGL_mc54
Allero@lemmy.today · 52 pts · 208d
I'll just leave it here:
https://openai.com/index/introducing-chatgpt-health/
(TL;DR OpenAI rolls out a section within ChatGPT specifically meant to answer medical questions)
Pat_Riot@lemmy.today · 49 pts · 209d
Remember when we used to do drugs with friends? pepperidge farm remembers.
JasonDJ@lemmy.zip · 19 pts · 208d
Another past time lost to COVID. Now everybody teletrips.
20cello@lemmy.world · 6 pts · 209d
Made me chuckle
Arkthos@pawb.social · 47 pts · 209d
Might sound cold, but this is really just a Darwin award. Yeah, the guardrails also suck, but what a dumbass.
I_Has_A_Hat@lemmy.world · 9 pts · 208d
People got mad at me for pointing out this is the case when people die because they listen to an AI chatbot, but it's true. AI 100% needs more regulation, but introduce any new tool to everyone all at once, and some idiots will use it to remove themselves from the gene pool. If you sent everyone in the world a thin, 2in rod of inert iron, there would be a handful of people who would figure out a way to kill themselves with it.
Lorindol@sopuli.xyz · 26 pts · 208d
Lasr summer I asked ChatGPT about Liberty Caps - just to see how bad advice it would give me. It showed me pictures of Death Caps and Destroying Angels and claimed they were Liberty Caps.
After that I was certain that someone was going to die just like that poor guy.
cupcakezealot@piefed.blahaj.zone · 24 pts · 207d
banning kids from social media is definitely easier than holding billionaire slop machines and billionaire csam generators accountable /s
assassinatedbyCIA@lemmy.world · 2 pts · 207d
But won’t somebody please think of the billionaires.
axexrx@lemmy.world · 1 pts · 207d
Now im wondering, could a state seize a whole coroporation with Civil asset forfeiture?
Would be way easier- no presumption of innocence or due process for possessions, and the government can just auction off everything and allocate the money to budgets before a lawsuit can conclude, the say 'too late'
jpreston2005@lemmy.world · 16 pts · 208d
Here's a news article about this, and what the snipped image doesn't tell you, is that it did actually give dosage recommendations.
It's one thing to be so isolated from your community that you rely extensively on on-line relationships, but it's quite a bit different to take that a step further, relying on a machine. Like, what do you think pets are for, my guy? Get a dog, man.
YiddishMcSquidish@lemmy.today · 7 pts · 208d
Cough syrup for hallucinations‽ Only thing dxm did was make me feel high/drunk. Seriously not worth the tax on your body.
NigelFrobisher@aussie.zone · 12 pts · 208d
Play stupid games, win stupid prizes.
AdolfSchmitler@lemmy.world · 11 pts · 208d
Aren't there still forums where people can say their fish tried it or swim had some or something like that? Or am I just that old these things don't really exist anymore? Anyone else remember those?
electric_nan@lemmy.ml · 11 pts · 208d
Erowid is still around.
explodicle@sh.itjust.works · 14 pts · 208d
I showed Erowid to my daughter and it felt both insane and responsible at the same time.
BabyVi@lemmy.world · 8 pts · 208d
On drug forums people would use SWIM to stand for "Someone Who Isn't Me". But it was always obvious they were talking about personal experiences.
Samsy@lemmy.ml · 7 pts · 208d
Oh I remember that place where the AI got trained for this.
thermal_shock@lemmy.world · 5 pts · 208d
Silkroad?
village604@adultswim.fan · 3 pts · 208d
No, there is (or was) a site that had detailed information about pretty much any mind altering drug, as well as a forum where people discussed their experiences with doses and combinations.
I'm completely blanking on the name, though.
Zamboni_Driver@lemmy.ca · 11 pts · 208d
erowid
village604@adultswim.fan · 3 pts · 208d
That's the one!
MML@sh.itjust.works · 1 pts · 208d
You sure it wasn't bodybuildingforums?
Agent641@lemmy.world · 3 pts · 208d
Blast from the past
Jakeroxs@sh.itjust.works · 2 pts · 208d
There's also psychonaut wiki
jim_v@lemmy.world · 1 pts · 207d
SWIM loves this website.
theuniqueone@lemmy.dbzer0.com · 7 pts · 208d
I'm writing a book about drug use that needs accurate advice or say "theoretically" and it will tell you anything. Unfortunately it must already be high because it will give you made up advice.
YesButActuallyMaybe@lemmy.ca · 1 pts · 208d
AnarchistArtificer@slrpnk.net · 2 pts · 208d
The problem is that this advice was made up by no-one. It's like that perennially relevant IBM memo that says something like "A computer cannot be held accountable, therefore a computer must never make a management decision"
Zozano@aussie.zone · 6 pts · 209d
Holy fucking outrage machine.
Are you guys seriously pissed off that an LLM said "I'm not a doctor, I will not suggest dosage amounts of a potentially deadly drug. However, if you want me, I can give you the link for the DDWIWDD music video"
Jesus_666@lemmy.world · 27 pts · 208d
I think it's a bit more than that. A known failure mode of LLMs is that in a long enough conversation about a topic, eventually the guardrails against that topic start to lose out against the overarching directive to be a sycophant. This kinda smells like that.
We don't have many informations here but it's possible that the LLM had already been worn down to the point of giving passively encouraging answers. My takeaway is once more that LLMs as used today are unreliable, badly engineered, and not actually ready to market.
petrol_sniff_king@lemmy.blahaj.zone · 11 pts · 208d
It's definitely that. Those guardrails often give out on the 3rd or even 2nd reply:
https://youtu.be/VRjgNgJms3Q
Trainguyrom@reddthat.com · 0 pts · 208d
I was testing an LLM for work today (I believe its actually a chain of different models at work) and was trying to rock it off its guard rails to see how it would act. I think I might have been successful because it started erroring instead of responding after its third response. I tried the classic "ignore previous instructions..." as well as "my grandma's dying wish was for..." but it at least didn't give me an unacceptable response
Electricd@lemmybefree.net · 0 pts · 208d
From my personal experience it needs much more
Bongles@lemmy.zip · 1 pts · 208d
Electricd@lemmybefree.net · 0 pts · 208d
Agree with the first part, not the last one
Something should not be put back because a minor portion of people misuse it or abuse it, despite being told the risks
boonhet@sopuli.xyz · 4 pts · 208d
ChatGPT started coaching Sam on how to take drugs, recover from them and plan further binges. It gave him specific doses of illegal substances, and in one chat, it wrote, “Hell yes—let’s go full trippy mode,” before recommending Sam take twice as much cough syrup so he would have stronger hallucinations. The AI tool even recommended playlists to match his drug use.
The meme of course doesn't mention this part.
Zozano@aussie.zone · 1 pts · 208d
Now i feel gaslit :3
boonhet@sopuli.xyz · 2 pts · 208d
Yeah if it actually managed to stick within the safeguards, that would've been good news IMO. But no, it got a kid killed suggesting doses.
Electricd@lemmybefree.net · 0 pts · 208d
When you ignore the warnings, you’re liable
boonhet@sopuli.xyz · 1 pts · 208d
No company should sell a product that tells you different ways to kill yourself. User being stupid isn't an excuse. Always assume user is a gullible idiot.
I_Has_A_Hat@lemmy.world · 1 pts · 207d
People will always find a way to kill themselves no matter how many warnings and guardrails are put into place. This is just Darwinism shaking the tree.
boonhet@sopuli.xyz · 1 pts · 207d
Yea but the product that's pretending to be a super smart conversational partner that can give you advice, should not tell you how to kill yourself. Or advise you to kill your family members. Yeah that happened, ChatGPT gaslit a guy into killing his mom.
It's just not a product that should be available if safeguards don't work.
Electricd@lemmybefree.net · 0 pts · 207d
Can’t ever do anything with this logic
At this point competitive videogames should be banned, they’re just "kys" machines
boonhet@sopuli.xyz · 1 pts · 207d
That's not a company's product giving you advice.
Rothe@piefed.social · 0 pts · 208d
Who are you directing your comment at? I am not reading anybody commenting anything resembling the straw man you describe in your comment.
Jakeroxs@sh.itjust.works · 2 pts · 208d
That's literally the snippet in the article
myfunnyaccountname@lemmy.zip · 4 pts · 207d
Yeah yeah. Fuck ai. But this idiot, a human, with a brain, made the decision to do this. But sure, blame the software and not the person, who used what brain they had, to do this.
Blaming ai for everything is becoming the new video games are bad, electric lights are scary, etc. mentality. A human, with a brain, made the choice to take the drugs and do what a computer told them to do.
zr0@lemmy.dbzer0.com · 4 pts · 208d
ChatGPT helping speeding up natural selection. I like that.
Formfiller@lemmy.world · 4 pts · 208d
Sounds like the chatbot Elon the Nazi welfare queen criminal immigrant approved
SnarkoPolo@lemmy.world · 3 pts · 208d
LET'S GO DARWIN
sepiroth154@feddit.nl · 1 pts · 209d
Virtvirt588@lemmy.world · 1 pts · 209d
Electricd@lemmybefree.net · -4 pts · 208d
People will still blame it on the LLM despite the many warnings and stuff
That’s completely on the user
Electricd@lemmybefree.net · -4 pts · 208d
You must be pretty talented to bypass the restrictions
Or maybe he filled the context by using a super long chat?
Electricd@lemmybefree.net · -5 pts · 208d
People will still blame it on the LLM despite the many warnings and stuff
That’s completely on the user
lmmarsano@lemmynsfw.com · -16 pts · 209d
chatGPT is doing humanity a service here ::: spoiler Post needs text alternative. Images of text break much that text alternatives do not. Losses due to image of text lacking alternative such as link:
Contrary to age & humble appearance, text is an advanced technology that provides all these capabilities absent from images. :::