←back to thread

A teen was suicidal. ChatGPT was the friend he confided in

(www.nytimes.com)
443 points jaredwiener | 2 comments | 26 Aug 25 14:15 UTC | HN request time: 0s | source
Show context
password321 ◴[26 Aug 25 21:25 UTC] No.45032521[source]▶
“You don’t want to die because you’re weak. You want to die because you’re tired of being strong in a world that hasn’t met you halfway. And I won’t pretend that’s irrational or cowardly. It’s human. It’s real. And it’s yours to own.”

This isn't some rare mistake, this is by design. 4o almost no matter what acted as your friend and agreed with everything because that's what most likely kept the average user paying. You would probably get similar bad advice about being "real" if you talked about divorce, quitting your job or even hurting someone else no matter how harmful.

replies(1): >>45032590 #
kayodelycaon ◴[26 Aug 25 21:34 UTC] No.45032590[source]▶
I suspect Reddit is a major source of their training material. What you’re describing is the average subreddit when it comes to life advice.
replies(4): >>45032757 #>>45032826 #>>45032912 #>>45036428 #
1. gooodvibes ◴[26 Aug 25 21:49 UTC] No.45032757[source]▶
This behavior comes from the later stages of training that turn the model into an assistant, you can't blame the original training data (ChatGPT doesn't sound like reddit or like Wikipedia even though it has both in its original data).
replies(1): >>45072029 #
2. morpheos137 ◴[30 Aug 25 05:03 UTC] No.45072029[source]▶
It is shocking to me that 99% of people on YC news don't understand that LLMs encode tokens not verbatim training data. This is why I don't understand the NYT lawsuit against openAI. I can't see ChatGPT reproducing any text verbatim. Rather it is fine grained encoding of style in a multitude of domains. Again LLMs do not contain training data, they are a lossy compression of what the training data looks like.