Hugh 2

[Hugh 2 @Hugh_42]

@GaryMarcus I don't think anybody actually misunderstands what p(doom) means.
Yud, recently: https://t.co/v6gKI2YEws
Me, 3 years ago: https://t.co/e4PyRHchpD

[Hugh 2 @Hugh_42]

@shakoistsLog Unknown unknowns. We don't have past experience of doom to gather statistics about. There's a term for it: Knightian uncertainty
https://t.co/obBaKdFYkC

[Hugh 2 @Hugh_42]

@So8res If a definitely-not-an-agent oracle is publicly available, won't somebody just query it in a loop, feeding (some of) its output back to its input, effectively creating an agent anyway?

[Hugh 2 @Hugh_42]

@tombibbys I think you can appeal. Isn't using excerpts for commentary purposes "fair use"?

[Paul Graham @paulg]

The reason so many people look for an ulterior motive for the AI labs asking to be regulated is that they don't grasp that models could be dangerous. But if you try assuming models are getting dangerous, or at least unpredictable, everything falls into place.

[Hugh 2 @Hugh_42]

@GaryMarcus It sounds like their motives are good. "It's four corporations agreeing not to compete on the one thing regulators were about to force them to adopt — safety standards."
https://t.co/ukYkAyOzKU

[Cheyenne Hunt @CheyenneHuntCA]

Today I can publicly announce that I am part of a landmark lawsuit against all of the major AI companies. I'm suing Anthropic, OpenAI, X, and Google because we deserve real AI safety standards, not shady deals cut by an unaccountable cartel of billionaires in a back room.

Since July, these companies have been meeting privately to plan a response to growing public pressure for real AI regulation. Then Anthropic's CEO published an essay saying the industry should slow down. Within hours, Musk and Altman said publicly that they agreed.

They're branding it as a safety breakthrough, but it's not. It's four corporations agreeing not to compete on the one thing regulators were about to force them to adopt — safety standards. This is a blatant violation of the Sherman Act, which outlaws corporations joining forces to dodge public pressure together instead of facing it alone. 

The timing isn't a coincidence, either. This "deal" was struck at the exact moment lawmakers were gaining real momentum to pass binding AI regulations. So instead of enforceable law, we got a deal cut among the companies themselves — one they wrote, they enforce, and they can rewrite whenever it’s in their best interest.

This publicly announced pinky promise gave Mike Johnson cover to adjourn the House of Representatives just before they could pass binding legislation on the topic.

If these companies actually believe their products could risk human extinction, how to move forward is not a decision four CEOs get to make behind closed doors, weighing their profit margins against the harms done to users, the economy, and the environment. That's the government's job — with real transparency into how the rules are written and how they are enforced. 

So I'm suing. Because rules written by the industry, for the industry, policed by the industry aren't safety standards. They're a grift. And the public should be writing the rules — not four billionaires.

[Hugh 2 @Hugh_42]

@Aella_Girl @michaelshermer ... "merit" referring to the hypothesis that the AI companies are not being straightforward

[Hugh 2 @Hugh_42]

@Aella_Girl @michaelshermer "x-risk is beneficial hype" was always bullshit, but "we need to slow down before we run out of money, look here's a good cover story" might have merit

[Hugh 2 @Hugh_42]

@ilex_ulmus Wouldn't you be happier living in the Holly Elmore medianworld?

[Hugh 2 @Hugh_42]

@robertskmiles @allTheYud Wait, if an idea is really dangerous I don't think you should even hash it. You'd have to type it at least once, and store the plaintext if you wanted to reveal it later (unless you can remember it character-for-character), and neither computers nor hashes are perfectly secure

[Hugh 2 @Hugh_42]

@robertskmiles @allTheYud Have you revealed any of your previously-hashed plaintexts yet?

[Hugh 2 @Hugh_42]

@allTheYud @tszzl I suspect a non-agentic superintelligence could wipe out its creator species in response to an unlucky (or malicious) prompt, then sit there waiting for the next prompt... until the power fails

[Hugh 2 @Hugh_42]

@paulg @KingRomstar This makes no sense to me. If they're secret, they're not doing any business yet. If they're "growing at 57% a week", they're not secret!

[Hugh 2 @Hugh_42]

@bryan_johnson I wish that too, but wishing for a thing does not make it so. I'll give you some links for further reading:
https://t.co/th0HSSQLy2
https://t.co/VdFWTqF8fT "Why advanced AI isn't like other technologies"
https://t.co/Y2mRkK8eSq

[Hugh 2 @Hugh_42]

@GremlinGrobber @ollkorrect0 She puts it in quotes as if it's not a real color. But chrome yellow and chrome colorless are recognized colors (gold and silver), why not chrome orange?

[Hugh 2 @Hugh_42]

@jf_saltarin @robertskmiles Replace “murder” with “delete” and you might see what he’s really talking about

[Rob Miles @robertskmiles]

An important principle:
Never pay someone to remove a problem that they themselves created

[Hugh 2 @Hugh_42]

@DanielleFong "On the internet, nobody knows you're a dog"
You can't confidently establish truth about the real world just from internet text.
On another level, how do you know you're not in a virtual machine with all inputs being controlled (you probably are). An AI can't be sure of anything

[Hugh 2 @Hugh_42]

@DanielleFong This is your caption to every food pic... must be a reference to something I am unaware of?

[Hugh 2 @Hugh_42]

@aliceisplaying @Aella_Girl I came here to say this

[Hugh 2 @Hugh_42]

@unsoldwill @So8res @allTheYud @KeyTryer If that worked, it would only be until the next person spoke to him