It gets worse the longer that you engage with the chatbot. OpenAI didn't expect for conversations to last for months and months, across thousands of messages. Of course, when they did learn that people were engaging with ChatGPT in this way, and that it severely compromised its already insufficient safeguards, their response was "huzzah, more engagement. How do we encourage more people to fall into this toxic cycle?"

It's the same level of "hallucinations" as always, that is, zero.

This isn't hallucinating (LLMs don't have a mind, they aren't capable of hallucinating, or any other form of thought), this is working as intended.

These things will tell you whatever you want to hear, their purpose isn't to provide information, it's to create addiction, to keep the customer engaged and paying as long as possible, regardless of consequences.

The induced psychosis and brain damage is a feature, not a bug, since it makes the victim more dependent on the LLM, and the cartel selling access to it.

Given the costs, and the amount of money already burnt building them, these companies need to hook as many people as possible as fast as possible, and get them addicted enough that when they raise the prices 100X to a sustainable level their victims won't be able to leave.

And they need to do this fast, because the money is running out.