Conversation

Is the answer to AGI “alignment”, which implicitly means centralized control by the few people that it’s aligned with? Or is the answer more like decentralization, to ensure that everyone has access rather than a privileged few?
Quote Tweet
There's a third camp: - a group that worries the other two groups will sow enough fear about AI that it becomes the sole province of a few large, centralized incumbents, with the rest of us left at their mercy. AI safety == I am safe from BigCos b/c I have my own AI. twitter.com/atroyn/status/…
Show this thread
46
381
Replying to
AI notkilleveryoneism (formerly known as "alignment") is about the empirical, falsifiable, statement-about-reality that alignment to anything or anyone is difficult. Getting it for even one AI out of a group is difficult. Multiple unaligned AGIs kill you just as fast.
9
144
Replying to
Why do they kill you rather than the other tribe who is not providing them necessary resources (compute, etc)? How do they kill you without resources from their tribe? In terms of empiricism, AGIs don't exist — but we are seeing LLMs customized for different value sets already.
6
8
Replying to
Very very smart things can rapidly render themselves independent of human infrastructure. There's been thorough analysis of (very primitive) nanotech, but leaving that aside, you could imagine sufficiently advanced biotechnology to build servitor shoggoths.
3
39
Replying to
I’m skeptical of the hard takeoff nanotech scenario as (a) nanotech hasn’t actually been invented yet, suggesting serious practical roadblocks and (b) there are good arguments that Drexler-style nanobots are actually not physically possible.
Quote Tweet
See also Drexler/Smalley, where Nobel Laureate Richard Smalley argues that molecular assemblers of the type Drexler described aren't actually physically possible. en.wikipedia.org/wiki/Drexler%E
Show this thread
2
13
But very few people would have read Drexler’s papers, and so most of the simply believed the assertion that he hadn’t addressed the issues. They had no direct knowledge to the contrary, and no informed reason to doubt an eminent expert who wasn’t telling the whole story.
1
1
Show replies
Replying to and
You two aren't talking to each other. You're smarting at each other. You're both brilliant enough to be able to do something more clever & glorious than Memetic Repeater v. 3.8, Homegrown Edition™. Please, please consider being deeper than this with each other.
1
3