I literally actually don't know what you mean by this. What in this sense distinguishes a human from an LLM using chain-of-thought?
@esyudkowsky
-
Training Large Models Has Become Trivially Easy in 2023
By
–
Sir, this is 2023. Training over a hundred million parameters is what happens when you press the "train Lora" button on the HTML interface.
-
AGI Safety Victory: Ensuring Beneficial Superintelligent Expansion
By
–
I can't quite say that I'd declare victory, because I need to know what use is going to be made of the rest of the accessible galaxies, if they all turn into cities of sapient life having fun. But that those AGIs were at least safe for humans–I am happy to call that game won.
-
ASI Competition for Resources Threatens Human Survival
By
–
We compete for matter, negentropy, and humans creating actually-competitive additional ASIs if we're allowed to stick around. Or more plainly, if an ASI boils Earth's oceans as coolant for computation, humanity doesn't survive that.
-
The Babysitter: A Masked Hero Who Regenerates Mid-Episode
By
–
Concept: Children's show with a masked hero, The Babysitter; whose voice, height, and face outside the mask, sometimes randomly change mid-episode. Then: Anyone with that mask will be recognized by your kids as one of The Babysitter's regenerations, not an alarming stranger.
-
Red-teaming Grok to produce corporate platitudes
By
–
I wonder how much work it will be for red-teamers to get Grok to spout blank-faced corporate pablum.
-
Current AIs are Five-Year-Olds: Restrict Their Access Rights
By
–
I say again: Current AIs are five-year-olds. Do not give them read or write permissions to anything important, especially if they have literally any exposed attack surface (such as reading externally created text). https://
t.co/JtyfF0HJ74 -
Prediction Markets for Waymo Human Intervention Monitoring
By
–
I decided that I don't want to monitor this issue hard enough to adjudicate a prediction market myself; but making the prediction market for whether Waymo is relying on amounts of human intervention near that of Cruise, would be one way to promote rumors to public best-guesses.
-
Deontology and computational efficiency in moral decision-making
By
–
It's meaningless to speak of deontology being computationally cheap, except I suppose in the same way that being a rock as cheap, without it being the case that deontology is doing some task cheaply. That task, or target, is mapping preferred outcomes onto actions. Deontology
-
Stupid AI Using Computationally Cheap Shortcuts Relative to Humans
By
–
Sure, but not the same decisions where humans use computationally cheap shortcuts. If we can see them using a computationally cheap and lossy shortcut, where we can see the lossiness from our level, that means they're being stupid relative to us.