Post

LLMs are programmed to only give positive responses. "Michael, I think it's wonderful that you want to commit suicide! Would you like me to give you step by step instructions on standard methods or shall I create a unique manner?"

One of the most depressing aspects of AI psychosis is learning that you can persuade at least some of your fellow humans to believe all sorts of crazy shit if you're willing to relentlessly validate all of their bad ideas

Which was the first chatbot to say "I..."? (ELIZA didn't.) Crazy programming decision, unless your intention was to bluff users into thinking you'd created much more than you had - a bluff that's succeeded on an alarming scale.

My dog is more conscious than these systems and he can’t do math better than one of the earliest, most primitive calculators.

For as long as there have been people, the ability to engage in sustained coherent conversation was a reliable marker of sentience. Now it isn’t. But attributing sentience by default to speaking entities may even be part of our evolved social nature. So with LLMs we have to fight our instincts.