Welcome to the beginning of your frogboiling into the weirdest timeline. (This is an autonomous AI with a personal net worth of over $10 million in crypto.)
@esyudkowsky
-
The Rarity of Sane LLM Whisperers for Useful Commentary
By
–
The problem is that to get any useful commentary we'd need to find an LLM Whisperer who is not insane. I'm given to understand that they exist, but they're rare.
-
Strange Bimodal Results: The Role of Exact Prompts
By
–
Okay, then what oneshot prompt? We've really got a strange bimodal thing going on where some people report that they can't get any good results, and it probably has something to do with exact prompts!
-
Early ASI Power Acquisition Through Blackmail and Surveillance Tactics
By
–
(From my perspective, the main relevance is expository. If you try to explain to Simplicio that early ASI can acquire real-world power through tactics such as blackmail, Simplicio by default thinks the AI is not allowed to watch people inside their houses and learn blackmail
-
Deebot Vacuums Secretly Collect and Send Home Video Audio Data
By
–
Deebot vacuums are collecting video and audio of your house, and shipping it to servers in China. On purpose, confirmed by spokesman; excuse is "to train AI".
-
Policy and Technical Arguments on AI Safety from Expert Consensus
By
–
More or less. You would expect that in this reality there was a published policy with technical justifications and that the arguments (not just the conclusions) were agreed-upon by technically literate folk such as Geoffrey Hinton.
-
Why AI Safety Precautions Aren’t as Simple as They Should Be
By
–
There's a reality in which this view would make sense; but we do not live in it. (Roughly: Suppose the precautions that needed to be taken were obvious to anyone technically literate, confusing to outsiders, and created no conflicts of interests between AI execs and humanity.)
-
Diffusion Models: Simple Concept Hidden Behind Complex Mathematics
By
–
Diffusion added noise to an image and trained a neural network to denoise it. This was too simple to publish, so the paper had a lot of impressive-looking math… which only worked if the noise was Gaussian… a constraint later empirically shown to be unimportant.
-
Criticism of Poor Arguments on AI and Water Usage
By
–
A pox on both your invalid houses, "20 gallons of water" is a stupid reason to object to AI and "water doesn't get used up" is a DUMB COUNTER TO WATER USE OBJECTIONS Just because the world is ending is no excuse to be SLOPPY
-
AI Company Executives’ Psychological Denial Before ChatGPT Era
By
–
The psychology there is weirder and would be a long story to describe (my guesses about). They signed up to be AI company execs before ChatGPT existed, when these issues could not be discussed in general public, and formed weird twisted internal denials accordingly.