Toby Tremlett🔹

• Toby Tremlett🔹 23 Jul 2026 13:19 UTC

  4 points

  0 ∶ 0

  on: Hard-to-re­verse de­ci­sions de­stroy op­tion value

  I’m curating this post from 2017 because:
  a) ‘hard to reverse decisions destroy option value’ is a true and useful idea and
  b) this problem is relevant again as we consider the multiple ways the EA movement could develop, if Anthropic or OpenAI IPOs do change the funding environment.

• Toby Tremlett🔹 23 Jul 2026 8:44 UTC

  5 points

  1 ∶ 0

  on: So you want to start some­thing for an­i­mals: a map of the incubators

  Thanks Sofia! Not a direct answer to your post, but this is a good example of why I’m not currently planning on downranking AI-written posts. This post is a great resource, and the fact that it’s AI-assisted doesn’t really affect the value.

• Toby Tremlett🔹 22 Jul 2026 9:45 UTC

  2 points

  0 ∶ 0

  in reply to: James Herbert’s comment on: to­bytrem’s Quick takes

  Absolutely! Those are most of my questions, but I don’t think I have one about relative priority (i.e. are community-building teams spending too much/​ enough/​ not enough of their effort on marketing). I’ll add something.

• Toby Tremlett🔹 21 Jul 2026 19:18 UTC

  15 points

  0 ∶ 0

  on: to­bytrem’s Quick takes

  Me and @Franare interviewing @GraceAdams🔸 tomorrow evening. Is there anything you’d like us to ask her?
  
  

New? Start here.

• Toby Tremlett🔹 20 Jul 2026 13:21 UTC

  5 points

  1 ∶ 0

  in reply to: John Salter’s comment on: Mal­colmMcLeod’s Quick Takes

  Such as, for example, the Effective Altruism Forum :)

• Toby Tremlett🔹 16 Jul 2026 10:43 UTC

  5 points

  0 ∶ 0

  on: Count­ing an­i­mals: Stable pop­u­la­tion size is not equiv­a­lent to pri­or­ity level

  I’m curating this post.

  Firstly because it makes a useful and well backed up point—that stable population estimates are a misleading input into your analyses.

  Secondly because: We often extoll the values of pretty shallow fermi-estimate style analyses. This post is a helpful counter-weight against the BOTEC blasé: some inputs are uncertain enough that we should (a) investigate them further, even for a quick BOTEC and/​or (b) remember that the output of our analyses could change radically based on tweaking a few assumptions.
  
  Thanks Abraham and Mal!

• Toby Tremlett🔹 16 Jul 2026 9:02 UTC

  9 points

  0 ∶ 0

  on: to­bytrem’s Quick takes

  Just under a month until the Cluelessness Critiques Competition deadline. How’s it going? Any hints on what people are writing about?

• Toby Tremlett🔹 13 Jul 2026 8:24 UTC

  6 points

  0 ∶ 0

  in reply to: huw’s comment on: MacKen­zie Scott’s $26 billion, in QALYs: an in­ter­ac­tive, ev­i­dence-weighted CEA

  Actually, we don’t downrank them, we just auto-tag them as AI-generated (although this one hasn’t run automatically- doh)

• Toby Tremlett🔹 10 Jul 2026 22:23 UTC

  2 points

  0 ∶ 0

  in reply to: NickLaing’s comment on: to­bytrem’s Quick takes

  She had to delay so it’s not too late, but I already included that exact question :)

• Toby Tremlett🔹 10 Jul 2026 9:52 UTC

  5 points

  0 ∶ 0

  on: to­bytrem’s Quick takes

  I’m recording a podcast (TWCBB) with @Lauren Gilbert [edit: we had to reschedule so I can still collect questions]. If there’s anything you’d like us to ask her, comment here.

• Toby Tremlett🔹 10 Jul 2026 9:44 UTC

  4 points

  0 ∶ 0

  in reply to: huw’s comment on: AI 2040: Plan A [thread]

  My guess is that politician’s staffers read it, and also that it elevates the status of the authors such that they are likelier to be taken on to advise. I presume it’s less likely Daniel Kokatajlo would have been interviewed by Bernie Sanders if he hadn’t written AI 2027.

  The smart thing in the framing here is that politicians will soon (hopefully) be reaching for more comprehensive plans on what to do about AI. And this is called ‘Plan A’. So it’s not crazy to think that something like this may end up influencing legislation.
  

AI 2040: Plan A [thread]

• Toby Tremlett🔹 9 Jul 2026 12:45 UTC

  2 points

  0 ∶ 0

  in reply to: Austin’s comment on: akrols­mir’s Shortform

  Thanks Austin, I’ll pass this on (I’m guessing this is a refactor bug)

• Toby Tremlett🔹 9 Jul 2026 12:41 UTC

  11 points

  1 ∶ 0

  on: What would an an­i­mal-al­igned AI be al­igned to?

  I’m curating this because aligning AI for the benefit of animals is a valuable and under-discussed problem, this is a valuable investigation, and I’d love to see it discussed further.

• Toby Tremlett🔹 9 Jul 2026 8:04 UTC

  4 points

  0 ∶ 0

  in reply to: Thomas Kwa🔹’s comment on: METR Time Hori­zon 2.0—The bench­mark you’ve been wait­ing for

  Thanks for the flag, looks like pangram didn’t run for some reason, we’ll fix it!
  (The automatic labelling thing was only launched last week, so there will likely be bugs, I appreciate all flags a lot!)

• Toby Tremlett🔹 8 Jul 2026 9:17 UTC

  5 points

  1 ∶ 0

  in reply to: WilliamKiely🔸’s comment on: The first video from Giv­ing What We Can’s new chan­nel is out now!

  FWIW thumbnails and titles on YouTube are generally in a constant A/​B test, and my guess is it’d be better to just let that algorithm run than to take individual opinions into account. I.e. I’d be surprised if GWWC wasn’t already experimenting with the thumbnails and titles.

Alice and Mir­ror Alice [short story]

• Toby Tremlett🔹 6 Jul 2026 13:16 UTC

  3 points

  0 ∶ 0

  in reply to: dan.pandori 🔸’s comment on: The Learn­ing Trap: What Si­mu­lated Clue­less Agents Re­veal About the Unaware­ness Argument

  Hey Dan! In this case we have decided to leave this up and count you as eligible (to be clear, for the first stage, it could still be a desk rejection in the end) because this already has some valuable discussion. From now on we’ll ask people to take their posts down if they are posting and hoping to be included in the competition. I’ll also make sure that the announcement post is clearer. Cheers!

• Toby Tremlett🔹 6 Jul 2026 8:06 UTC

  6 points

  0 ∶ 0

  in reply to: Rohan Prasad’s comment on: Yu­val Harari: philan­thropy prin­ci­ples + 3 AI-fo­cused char­i­ties he recom­mends (7/​4, link ud­pate)

  I agree, seems very valuable to share, thank you!