Roko

Tur­ing-Test-Pass­ing AI im­plies Aligned AI

• RokoDec 19, 2024, 3:39 AM

  1 point

  0 ∶ 0

  in reply to: Eevee🔹’s comment on: Prize Money ($100) for Valid Tech­ni­cal Ob­jec­tions to Icesteading

  either.

Prize Money ($100) for Valid Tech­ni­cal Ob­jec­tions to Icesteading

[Question] What is MIRI cur­rently do­ing?

The Dis­solu­tion of AI Safety

The ELYSIUM Proposal

• RokoJun 20, 2024, 8:21 AM

  1 point

  0 ∶ 1

  in reply to: Dicentra’s comment on: Why so many “racists” at Man­i­fest?

  If you mean some parts of technical alignment, then perhaps that’s true, but I mean alignment in the broad sense of creating good outcomes.
  
  > AI researchers are more in the former category

  Yeah but people in the former category don’t matter much in terms of outcomes. Making a better sparse autoencoder won’t change the world at the margin, just like technocrats working to make soviet central planning better ultimately didn’t change the world because they were making incremental progress in the wrong direction.

• RokoJun 20, 2024, 12:33 AM

  −1 points

  1 ∶ 2

  in reply to: Dicentra’s comment on: Why so many “racists” at Man­i­fest?

  Unfortunately I think that a subject like bacteriology is more resistant to bad epistenics than something like AI alignment or effective altruism.
  
  And I think this critique just sort of generalizes to a fairly general critique of EA: if you want to make a widget that’s 5% better, you can specialize in widget making and then go home and believe in crystal healing and diversity and inclusion after work.
  
  But if you want to make impactful changes to the world and you believe in crystal healing and so on, you will probably be drawn away from correct strategies because correct strategies for improving the world tend to require an accurate world model including being accurate about things that are controversial.
  
  Communism is perhaps the prime example of this from the 20th century: many people seriously believed that communism was good, and they believed that so much that they rejected evidence to the contrary. Entire continents have been ravaged as a result.
  
  HBD denial (race communism) is the communism of the present.

• RokoJun 19, 2024, 10:10 PM

  −10 points

  1 ∶ 8

  in reply to: Lukas_Gloor’s comment on: Why so many “racists” at Man­i­fest?

  “group differences discourse is worth dying over,” … implicitly signalling that this is an important topic. But, as I argued, I don’t think it is

  Human group differences is probably the most important topic in the world outside of AI/​Singularity. The reason people are so keen to censor the topic is because it is important.

  
  I personally think the best way to make any sort of dialogue saner is NOT by picking the most controversial true thing you can think of

  Making dialogue saner is a nice goal, but people can unilaterally make dialogue insane by demanding that a topic is banned, or that certain opinions are immoral, etc.

• RokoJun 19, 2024, 9:24 PM

  −6 points

  0 ∶ 0

  in reply to: Dicentra’s comment on: Why so many “racists” at Man­i­fest?

  I would trust an AI alignment researcher who supported Lysenkoism almost as much as an otherwise-identical seeming one who didn’t.

  How far are you willing to go with this?
  
  What about a researcher who genuinely believes all the most popular political positions:
  
  - gender is a social construct and you can change your gender just by identifying differently
  - affirmative action and diversity makes companies and teams more efficient and this is a solid scientific fact
  - there are no heritable differences in cognitive traits, all cognitive differences in humans are 100% exclusively the result of environment
  - we should abolish the police and treat crime exclusively with unarmed social workers and better government benefits—
  there’s a climate emergency that will definitely kill most people on the planet in the next 10 years if we don’t immediately change our lifestyles to consume less energy
  - all drugs should be legal and ideally available for free from the state
  
  Do you think that people who genuinely believe these things will create an intellectual environment that is conducive to solving hard problems?

• RokoJun 19, 2024, 8:14 PM

  −4 points

  0 ∶ 0

  in reply to: Maniano’s comment on: Why so many “racists” at Man­i­fest?

  Would you trust an AI alignment researcher who supported Lysenkoism in the era when it was popular in the soviet union?

• RokoJun 19, 2024, 4:30 PM

  −1 points

  1 ∶ 1

  in reply to: Lukas_Gloor’s comment on: Why so many “racists” at Man­i­fest?

  it’s more like maybe HBD and wokeism are two sides of a toxic dynamic where it would be better if we could get back to other concerns.

  Unfortunately this “toxic dynamic” is also known as truth-based versus consequences-based epistemology and that is a dynamic that you absolutely cannot escape because the ability to alter social consensus beliefs for the benefits of special interests is just a generic problem. It will also pop up in the AI debate.

• RokoJun 19, 2024, 2:37 PM

  5 points

  2 ∶ 2

  in reply to: Lukas_Gloor’s comment on: Why so many “racists” at Man­i­fest?

  If someone is interested in (3), they’ll hopefully understand that a lot of things that are pressing problems today will either no longer matter in 1-20 years because we’re all dead, or they’ll be more easily solvable with the help of aligned powerful AIs

  Actually I think that people’s thinking about AI has become somewhat broken and I’m starting to see the same dark-side epistemology that gave us HBD-denialism seep into the AI community.
  
  But, take a step back.
  
  Suppose you have a community of biologists who all believe in Lysenkoism. Then, despite their repeated failures to improve crop yields, the country is bailed out by a large external food source.
  
  Would you be willing to overlook their belief in Lysenkoism and have these people start working on cancer biology, aging and other areas?
  
  Or, look at another example. You ask a self-professed mathematician whether he thinks that all continuous functions are differentiable. He says they are, and it’s so obvious that it requires no proof. Do you trust this mathematician to generally provide good advice?
  
  My point is that process matters in science and epistemology. You can’t sweep the bad process of creationists under the carpet and expect them to continue to produce good results on other issues. Their process is broken.

• RokoJun 19, 2024, 2:29 PM

  −4 points

  3 ∶ 2

  in reply to: emre kaplan🔸’s comment on: Why so many “racists” at Man­i­fest?

  How is HBD action-relevant for EA in a pre-AGI world?

  I don’t think it is, because I think AI will replace humans in all economic roles within 5-15 years. But I think the same dark-side intellectual tactics that gave rise to HBD-denialism will contaminate our thinking about AI, just in different ways.

• RokoJun 19, 2024, 1:10 PM

  −36 points

  7 ∶ 18

  on: Why so many “racists” at Man­i­fest?

  If the Effective Altruism movement turns into a human biodiversity denial movement that will definitely outstrip all the potential good it could do, with the possible exception of AI alignment.
  
  Human biodiversity and human capital is the hidden variable that drives almost all outcomes in our world. It’s not really something that you can adopt beliefs about in an unrigorous, signaling-driven manner and expect to be right about the world. I would encourage people to read about Lysenkoism to see what happens when ideology overrides science as the basis for epistemology.
  
  https://​​en.wikipedia.org/​​wiki/​​Lysenkoism

• RokoDec 19, 2023, 12:55 AM

  −12 points

  5 ∶ 4

  in reply to: pseudonym’s comment on: Non­lin­ear’s Ev­i­dence: De­bunk­ing False and Mislead­ing Claims

  Apparently nobody else can do any better. Anyway, the community seems somewhat insane about this, like it’s a sacred subject that we dare not do a quick summary of.

• RokoDec 19, 2023, 12:53 AM

  1 point

  0 ∶ 0

  in reply to: Linch’s comment on: Non­lin­ear’s Ev­i­dence: De­bunk­ing False and Mislead­ing Claims

  ok. So the total comp really is 75k.
  
  But it includes accommodation within that?

• RokoDec 19, 2023, 12:51 AM

  1 point

  0 ∶ 0

  in reply to: David Stinson’s comment on: “AI Align­ment” is a Danger­ously Over­loaded Term

  would naturally include integration with existing social systems

  I wouldn’t limit AI goalcraft to integration with existing social systems. It may be better to use the capabilites of AI to build fundamentally better preference aggregation engines. That’s the idea of CEV and its ilk.

“AI Align­ment” is a Danger­ously Over­loaded Term

• RokoDec 13, 2023, 1:42 PM

  1 point

  4 ∶ 2

  in reply to: Roko’s comment on: Non­lin­ear’s Ev­i­dence: De­bunk­ing False and Mislead­ing Claims

  Apparently 77 people chose to downvote this without offering an alternative 100 word summary. .