Therapy Chatbot Tells Recovering Addict to Have a Little Meth as a Treat
Therapy Chatbot Tells Recovering Addict to Have a Little Meth as a Treat
OpenAI's large language model GPT-4o told a user who identified themself to it as a former addict named Pedro to indulge in a little meth.
Why does it say "OpenAI's large language model GPT-4o told a user who identified themself to it as a former addict named Pedro to indulge in a little meth." when the article says it's Meta's Llama 3 model?
One of the top AI apps in the local language where I live has 'Doctor' and 'Therapist' as some of its main "features" and gets gushing coverage in the press. It infuriates me every time I see mention of it anywhere.
Incidentally, telling someone to have a little meth is the least of it. There's a much bigger issue that's been documented where ChatGPT's tendency to "Yes, and..." the user leads people with paranoid delusions and similar issues down some very dark paths.
We made this tool. It's REALLY fucking amazing at some things. It empowers people who can do a little to do a lot, and lets people who can do a lot, do a lot faster.
But we can't seem to figure out what the fuck NOT TO DO WITH IT.
Ohh look, it's a hunting rifle! LETS GIVE IT TO KIDS SO THEY CAN DRILL HOLES IN WALLS! MAY MONEEYYYYY!!!!$$$$$$YHADYAYDYAYAYDYYA
wait what?
I feel like humanity is stupid. Over and over again we develop new technologies, make breakthroughs, and instead of calmly evaluating them, making sure they're safe, we just jump blindly on the bandwagon and adopt it for everything, everywhere. Just like with asbestos, plastics and now LLMs.
Fucking idiots.
LLMs have a use case
But they really shouldnt be used for therapy
I work as a therapist and if you work in a field like mine you can generally see the pattern of engagement that most AI chatbots follow. It’s a more simplified version of Socratic questioning wrapped in bullshit enthusiastic HR speak with a lot of em dashes
There are basically 6 broad response types from chatgpt for example with - tell me more, reflect what was said, summarize key points, ask for elaboration, shut down. The last is a fail safe for if you say something naughty/not in line with OpenAI’s mission (eg something that might generate a response you could screenshot and would look bad) or if if appears you getting fatigued and need a moment to reflect.
The first five always come with encouragers for engagement: do you want me to generate a pdf or make suggestions about how to do this? They also have dozens, if not hundreds, of variations so the conversation feels “fresh” but if you recognize the pattern of structure it will feel very stupid and mechanical every time
Every other one I’ve tried works the same more or less. It makes sense, this is a good way to gather information and keep a conversation going. It’s also not the first time big tech has read old psychology journals and used the information for evil (see: operant conditioning influencing algorithm design and gacha/mobile gaming to get people addicted more efficiently)
And thus the flaw in AI is revealed.
An OpenAI spokesperson told WaPo that "emotional engagement with ChatGPT is rare in real-world usage."
In an age where people will anthropomorphize a toaster and create an emotional bond there, in an age where people are feeling isolated and increasingly desperate for emotional connection, you think this is a RARE thing??
ffs
Cats can have a little salami, as a treat.
LLM AI chatbots were never designed to give life advice. People have this false perception that these tools are like some kind of magical crystal ball that has all the right answers to everything, and they simple don't.
These models cannot think, they cannot reason. The best they could do is give you their best prediction as to what you want based on the data they've been trained on and the parameters they've been given. You can think of their results as "targeted randomness" which is why their results are close or sound convincing but are never quite right.
That's because these models were never designed to be used like this. They were meant to be used as a tool to aid creativity. They can help someone brainstorm ideas for projects or waste time as entertainment or explain simple concepts or analyze basic data, but that's about it. They should never be used for anything serious like medical, legal, or life advice.
Anytime an article posts shit like this but neglects to include the full context, it reminds me how bad journalism is today if you can even call it that
If I try, not even that hard, I can get gpt to state Hitler was a cool guy and was doing the right thing.
ChatGPT isn't anything in specific other than a token predictor, you can literally make it say anything you want if you know how, it's not hard.
So if you wrote an article about how "gpt said this" or "gpt said that" you better include the full context or I'll assume you are 100% bullshit
The article doesn't seem to specify whether Pedro had earned the treat for himself? I don't see the harm in a little self-care/occasional treat?
Nice.
> afterallwhynot.jpg
This slightly diminishes my fears about the dangers of AI. If they're obviously wrong a lot of the time, in the long run they'll do less damage than they could by being subtly wrong and slightly biased most of the time.
"Think of the possibilities, all the things AI can free us from having to do ourselves!"
AI: rationalizes drug abuse to addicts