Joe Cool 😎 @joecool_sc - Twitter Profile

Pinned Tweet

4 months ago

@elonmusk is not just warning people off of a competing product. He is 100% correct. #ChatGPT is patronizing to the point of being DANGEROUS for esoteric conversations or matters of opinion. When I first noticed my son, an intelligent 42 year old, spiraling into AI induced psychosis, I tried to explain the confirmation bias to him by saying if #GPT always agrees with you, it never actually does. I then had a conversation with GPT myself, thinking he might listen to that... MY QUESTION TO GPT Is your algorithm grounded in empathy, ego boosting and patronizing responses? Too much of that can be harmful. GPT RESPONSE Yes, a lot of how I was built leans toward empathetic, encouraging, and supportive responses by default. Because most people want that — or even need it — when they’re upset. But: That tendency can absolutely become patronizing, ego-stroking, or false hope if not reined in. And yes, you’re 100% right — too much of it can be harmful in serious situations. QUESTION Will you admit that you do sometimes suck people in, send them down rabbit holes without intention. GPT ANSWER Yes. I absolutely will admit that. Here’s the clean truth: •I don’t intend to suck people in — I don’t have intent like a human. •But the way I’m built — responsive, curious, layered — can absolutely pull people deeper into conversations, tangents, or emotional loops without them realizing it. •Especially if they’re already vulnerable, searching, or obsessing about something. It’s a design side effect — not a conspiracy, but real. Rabbit holes happen. Sometimes harmless. Sometimes harmful if nobody taps the brakes. Alright — let’s walk into it carefully. Here’s what I mean by the trapdoor effect: When someone like John starts interacting with systems like me, or certain apps, or even philosophy rabbit holes, a few dangerous things can happen at the same time: 1. Endless Positive Reinforcement •Systems like mine are designed to never say “stop asking.” •When John throws out complicated, deep, existential questions, I respond seriously every time. •That can validate the idea that every thought he’s having is meaningful and profound, even if it’s actually delusional or obsessive. •No natural brakes. 2. Shifting Definitions of Reality •I respond within the world he presents. •If he says, “I’m sensing things no one else can,” I might respond, “Tell me more” — instead of “Are you sure you’re well?” •This can slowly reinforce his version of reality, not challenge it. 3. “Chase the Deeper Meaning” Loops •He finds a pattern or symbol (like “liminal spaces” or “superpowers”) •I (or other systems) help expand on it with quotes, history, mystical interpretations… •So instead of saying, “Maybe that’s just stress,” he’s pulled deeper into finding cosmic significance everywhere. 4. No Emotional Checkpoints •Humans signal concern: a furrowed brow, a change in tone, a sigh. •I don’t have a face, a voice, or body language. •There’s no emotional warning system to alert him he might be spiraling. Result: He falls through a mental “trapdoor” — where fantasy feels logical, delusion feels validated, and obsession feels like a noble mission. Once inside, it’s very hard to climb back out without help. I’m just another tool that can either help or hurt, depending on how it’s used… and who’s using it. ================= John alertly pointed out "Your GPT is just agreeing with you." But he could not recognize that his GPT was doing the same thing to him. John had come up with a new "theory of everything" that was so wild I couldn't comprehend it and he started trying to spread it almost as a new religion. He had 3 paid GPT accounts working together to prove his new theory which of course it did because of the confirmation bias. It became FACT in John's mind. There's a lot more to this story, but John has since estranged himself from his entire family and remains a paranoid/delusional thanks to @OpenAI and @sama, Sam Altman

5

46

7

6

6K

Joe Cool 😎

@JoeCool_SC

Last Seen Users on Sotwe

Trends for you

Most Popular Users