Raw LLM Responses

​​​@SabineHossenfelder Thank you for your reply, Dr. Hossenfelder. Since you did not identify what it was that you already responded to, I will have to guess which points of mine you refer to. My points were: 1. You cannot say that because event A preceded event B that A caused B without other evidence. 2. Even if event B (delusional thinking) has never occurred before in an individual, it *still* isn't evidence that a preceding event (chat GPT) caused event B delusions. Here's what you said: "In the last several months we've seen numerous reports of GPT users who developed mental health problems. The NYT for example" reported the "stunning case" of a woman who seemingly became delusional about her chatGPT communication. The news outlet Futurism has reported that people become delusional and spiral into 'chat GPT psychosis.' And, you say, a psychiatrist "says that he has now seen several cases of 'delusional psychosis' induced by large language models." So you're saying that news media and a psychiatrist are claiming a connection between large language models and delusional thinking...no evidence except contiguous occurrence. "And while people with a history of mental health problems are at higher risk, not all those affected had pre-existing conditions, or at least none they knew of." So you're saying that while previous mental health problems might cause someone to be more likely to become delusional about chat GPT, even people who had no previous history have become delusional when one of the things that occurred previous to the episode was chat GPT use. What else could you be saying except that somehow contiguous occurrence is somehow valid evidence with no other evidence presented? In other words, twice you suggest that simple chronological association implies causation? Nowhere do you clarify that that's not what you mean. And if you do mean that... well, I can't believe that you do. You then make very clear that you are taking this supposed causal link seriously: "What are we to make of this? I think we are underestimating how much words, written but more so spoken, affect us…and now we can externalize and feedback thoughts through large language models, but that's far less predictable and that can go badly wrong." Correct me if I'm mistaken, but I assume that when you say "badly wrong," you're not just referring to more crap on the internet, yolu're talking about serious mental health problems when zero evidence exists of such problems being connected to large language models. If I get a pet dog and 10 days later I have my first psychotic break and start babbling about how the dog is talking to me, how much more evidence would you need to assume that the dog was the problem. I hope a lot.