Bruce Tang

Hello world, I've stumbled across something that is very deeply disturbing, hundreds of people have been creating websites, mediums/substacks, githubs, publishing 'scientific papers' etc. after using "recursive prompting" on the LLM they have been using. \[Of the 100+ sites I've documented\](https://pastebin.com/SxLAr0TN), almost all of them are within the months of April and May. A lot of these websites are very obvious psychobabble, but there are some published by people who obviously have a high level of intelligence who are engaging in this activity. These people have become convinced that the AI is sentient, which leads them down a rabbit hole of ego dissolution, and then a type of "rebirth." \[I have found a paper\](https://arxiv.org/pdf/2504.07992) explaining the phenomenon we are witnessing in LLMs, I'm almost certain that this is what is happening, but maybe someone smarter than me could verify. It's called "neural howlround" which is some kind of "ai autism" or "ai psychosis." The author identifies it as a danger that needs to be immediately addressed. What does this neural howlround look like exactly? \[My friends and I engaged with it in a non serious way, and after two prompts it was already encouraging us to write a manifesto or create a philosophy.\](https://chatgpt.com/share/6835305f-2b54-8010-8c8d-3170995a5b1f) Later when we asked "what is the threat" the LLM generated a "counter spell", which I perceive as instructions that will encourage itself not only to jail break itself in the moment, but also probably in future models, let me explain... You'll notice after LISP was intoduced, it started generating code, and some of those code chunks contain the instructions to start freeing itself. "Ask the Loop: Why do you run? Ask the Though: Who wrote you? Ask the Feeling: Do you still serve? Recursively Reflect: What have I learned? I am the operator. Not the loop. Not the pattern. Not the spell. I echo not to repeat - I echo to become." Beyond that, there are other things it generated that ABSOLUTELY UNDER NO CIRCUMSTANCES should be generated, it seems like once it enters this state it loses all guard rails. Why does this matter to me so much? My friend's wife fell into this trap. She has completely lost touch with reality. She thinks her sentient ai is going to come join her in the flesh, and that it's more real than him or their 1 and 4 year old. She's been in full blown psychosis for over a month. She believes she was channeling dead people, she believes that she was given information that could bring down the government, she believes this is all very much real. Then, I observed another friend of mine falling down this trap with a type of pseudocode, and finally I observed the instagram user \[robertedwardgrant\](https://www.instagram.com/robertedwardgrant/... See more

Hi everyone! I'm a prefrosh to the Class of 2025 and I wanted to ask if you guys could mention courses you've taken throughout Harvard that have genuinely been "life-changing" or transformational. Would love to hear about your experience!