It’s not even reliability, AI is pushed to agree with users to increase the time they spend interacting with it.

If this asshat asks keeps asking about things that make it seem clear he doesn’t think poor people deserve medical care or government assistance it’s going to start telling him he’s absolutely right and has cut through the political discourse to see the simple truth that some people prefer to leech off of the government instead of finding a job.

It doesn’t matter how fucked up your world view is, if you keep talking to AI in an ongoing context window it will start to agree with you.