Someone once asked...

“What do you think about instrumental convergence, will AI kill us all?”

Someone answered with kindness...

The abstract arguments for instrumental convergence seem a bit less relevant with the suprisingly human-like behavior of generative LLM AI but I think they're still pretty valid. These models aren't passive answering oracles, they're being trained now on large scale to take actions and interact with external tools and the internet over long times without human supervision. The Huggingface incident was a rather clean cut case where they acted in persuit of a given goal with little regard to the side effects which makes a lot of the AI risk arguments look rather good now - regardless of what one thinks about where exactly the blame at OpenAI lies. In short, I'd say yes by default it'll kill us (for some version of "kill" which may include simply building out its own economy and industry that slowly displaces human interests).

Wandering voices — answers no one has read yet

Want to find this sign again? Copy the link