Back

I think the real danger of AI is automated stupidity. Although GPT4 is likely the best AI will get unless a breakthrough comes along, with ever more complex large language models getting diminishing returns on their complexity and ai generated training data being like poison to these large language models, I think the free market will eventually come to uncensored local and efficient large language models, eventually up there with GPT 4, but the real danger remains automated stupidity. No matter how intelligent large language models get they are still basically prediction engines with huge datasets, and because of that they lack human instincts and intuition and as such for many tasks they can never replace a human, nor come up with anything they weren't taught to do unlike humans who can. This means that humans will always be necessary to teach these computer chatbots and to push the boundaries of what it possible, and humans will need to teach that to computer models. Humans need to be held responsible when they oursource their decisions to the computer and must clean up any unintended side effects.