Magical Thinking on AI

There are a lot of sci-fi inflected ideas parroted in the media that are the cause of this. For dramatic purposes the AI in sci-fi has to act as an autonomous living being just like a human with its own goals and desires.

This doesn’t map at all to LLMs which are a big blob of statistics that predict the next couple letters of conversations. You can add as many bajillion parameters as you want to the blob, the next token isn’t ever going to something outside the set that lets it take over a computer and hack the nuclear silos, it’s just not how it works.

Friedman's arguments sound like it's 2023 and everyone is freaking out about ChatGPT and the imminent doom of everything. I thought we had moved past that!