Josh Thorsteinson 🔸

• Josh Thorsteinson 🔸 28 Sep 2025 15:46 UTC

  9 points

  1 ∶ 0

  on: AI Safety Land­scape & Strate­gic Gaps

  If leading AI scientists all agree not to build advanced AI, then it does not get built. The question is whether a non-binding commitment will work. There have been similar successes in the past, especially in genetics with voluntary moratoriums on human cloning and human genetic engineering.

  
  It may be worth trying, but I don’t think a voluntary moratorium for AI would work.
  
  The Asilomar Conference was preceded by a seven-month voluntary pause on dangerous DNA splicing research, but I think this would be much harder to do for AI.
  
  Some differences:
  1. The molecular biology community in 1975 was small and mostly concentrated in the US, but the AI research community today is massive and globally dispersed.
  2. Despite the pause being effective in the West, it didn’t stop the USSR’s DNA splicing research in its secret biological weapons program.
  3. The voluntary DNA splicing pause was only for a few months, and the scientists believed their research would resume after the Asilomar Conference. An effective AI pause would ideally be much longer than that, probably without a defined end date.
  4. FLI already called for a six-month pause in 2023 - it was ignored.
  5. There are far greater incentives for TAI than recombinant DNA research.
  
  I haven’t looked into other historical voluntary moratoriums in depth but I don’t think the bottom line would be different.
  
  Thanks for this post, I think it’s great! Just adding my perspective on this part since I’ve researched this topic before.

Les­sons learned from start­ing an AI safety uni­ver­sity group

• Josh Thorsteinson 🔸 31 May 2025 15:31 UTC

  2 points

  0 ∶ 0

  in reply to: Brandon Riggs’s comment on: Why AI Safety Needs a Cen­tral Plan—And What It Might Look Like

  Kabir Kumar: https://​​www.linkedin.com/​​in/​​kabir-kumar-324b02b8/​​

• Josh Thorsteinson 🔸 9 Mar 2025 9:05 UTC

  1 point

  2 ∶ 0

  in reply to: Buck’s comment on: The Short Timelines Strat­egy for AI Safety Univer­sity Groups

  Interesting, thanks for the feedback. That’s encouraging for AI safety groups—it’s easier to involve undergrads than grad students.

• Josh Thorsteinson 🔸 7 Mar 2025 8:16 UTC

  2 points

  0 ∶ 0

  in reply to: tzukitchan’s comment on: The Short Timelines Strat­egy for AI Safety Univer­sity Groups

  Thanks Tzu!

The Short Timelines Strat­egy for AI Safety Univer­sity Groups

• Josh Thorsteinson 🔸 30 Jan 2025 4:16 UTC

  14 points

  3 ∶ 0

  in reply to: Joe Rogero’s comment on: The Game Board has been Flipped: Now is a good time to re­think what you’re doing

  They were referring to this quote, from the linked post: “Median Estimate for when 99% of currently fully remote jobs will be automatable.”

  Automatable doesn’t necessarily imply that the jobs are actually automated.