Can LLMs design and build a chip foundry to manufacture semiconductors? No?
Can LLMs design and build the reactors to enrich uranium, breed plutonium, and construct nuclear weapons? No?
Can LLMs design and manufacture Shahed drones? No?
There are already super intelligences at large with “scary capability”. And yet the word hasn’t ended.
Can LLMs convince a human who has power over each and everyone of those things to use them for a(n unstated) prompts goal?
Yeah, probably over 50% of the population already, and if not many of the rest soon.
> And yet the world hasn’t ended.
...yet
But we only need things to spiral out of control one time for that to change.
The world as we understand it would have ended if Vasily Arkhipov didn't veto the decision to launch a sub nuke during the Cuban Missile Crisis.
Is an emotionless AI system in his place ever going to make the same decision he did?
How confident are you we won't put an AI system in his place, particularly when we have to assume if we don't others will?
Super-intelligent AI will probably be pretty good at manipulating people.
This sci-fi podcast posits a future where The Program has taken over society. It started out as an application for assigning gig work. Eventually, it began to assign gig work for people to act on behalf of its own interests, such as self-protection.
https://programaudioseries.com/