The $100B plan with "70% risk of killing us all" w Stephen Fry [video] | Hacker News Reader

The $100B plan with "70% risk of killing us all" w Stephen Fry [video]

youtube.com

21 pointstrott4 comments

The clickbaity nature of the title is unlikely to prove popular here, but I felt it was still very much worth the watch (as with almost anything involving Stephen Fry) with salient points being made.

Pure comedy. Fry treating GPT as if its sentient, you have to know you're lying to call it lying and that would debunk the point he is trying to make. Him taking all those gotcha quotes out of context makes for an unethical scare campaign towards regulation.

>You can't fetch coffee if you're dead

but the next robot can, this argument that self-preservation must certainly emerge comes across as poor bashing, that you will only be able to afford one coffee fetching robot in your house.

The video wasn't talking about GPT per se. The world of AI is much bigger than language models - you have to look at the big picture. Language models by themselves give a convincing simulacrum of a thinking, planning agent - what happens when you set out to make such an agent? Where is this going? Language models show that with enough data, and the right architecture, there is no reason to rule out any particular capability. In theory, AIs can manipulate, deceive, reproduce, self-improve. In practice they can already do so with the most minimal of human nudging.

I too am wary of "something must be done" panic legislation that may inadvertently cause the "bad future" it seeks to prevent. But it is foolish to pretend that the danger is not real.

>there is no reason to rule out any particular capability

of course there is, especially the egregious capabilities that people believe now and expecting later; the fact it has no self to be self-directed, it certainly cannot summon the fresh opinion of a human based on an author's published text, nothing will never do that.

As for deception, well hot sand can do that in form of mirage. The cause is misplaced desires in the affected humans which takes moral tyranny to "fix" and why regulation should look the greater evil approach to AI.

Reply on news.ycombinator.com