Alexander Turner 🔸

Why I Left Google DeepMind

• Alexander Turner 🔸 28 Jul 2026 17:46 UTC

  −1 points

  0 ∶ 0

  in reply to: Fran’s comment on: EA sex­ual ha­rass­ment: Where’s the ev­i­dence we’re worse than baseline?

  Agreed. I wrote one of the linked comments and don’t have reason to believe EA is worse than other areas.

  there’s something strange about a community that advocates for reducing suffering and encourages moral ambition retreating into “well, what’s the baseline?” when sexual harassment comes up.

  Agreed. “Let’s take all efficient moves to improve this bad situation” would be a nice response to hear, rather than “can you prove it’s worse than other places though?”. If Nathan’s argument is “there aren’t efficient moves left to take, we’ve entered diminishing returns” (as gestured at by “weigh the tradeoffs”), I think he should directly make that argument instead.

  ETA: Here is my full comment that Nathan quoted. If anyone has a social-economics rebuttal, I’d be interested in hearing it. Whether or not EA is doing better than other communities, the dynamic I pointed out is bad:

  I’ve repeatedly encountered the harmful notion that it is easy for victims to speak up; that they are immediately rallied behind, celebrated, and believed. This is not true. I was not rallied behind. I had to put in an unspeakable amount of effort for more than a year, and eventually invoke outside legal counsel, to see what I consider basic recognition.

  I’ve heard this from female friends, I’ve read about this (e.g. in the TIME article), and I myself paid social costs in sharing a different kind of negative experience. Victims often pay significant social costs to talk about their experiences.

  Community norms should not impose costs on sharing such information. I’m sorry you had to pay these costs, Frances. Thank you for speaking out. Hopefully this post decreases the cost in these communities. In fact, such important information should be socially subsidized, not taxed (since e.g. speaking out often requires reliving trauma, which is unpleasant; and most of the benefit is external).

• Alexander Turner 🔸 4 Mar 2026 14:59 UTC

  20 points

  14 ∶ 2

  on: CEA’s re­sponse to sex­ual harassment

  I’ve repeatedly encountered the harmful notion that it is easy for victims to speak up; that they are immediately rallied behind, celebrated, and believed. This is not true. I was not rallied behind. I had to put in an unspeakable amount of effort for more than a year, and eventually invoke outside legal counsel, to see what I consider basic recognition.

  The EA community has a significant undersupply of information from victims of abusive conduct, since the victims are often branded as “triggered” or “irrational”. I’ve heard this from female friends, I’ve read about this (e.g. in the TIME article), and I myself paid social costs in sharing a different kind of negative experience. Victims often pay significant social costs to talk about their experiences.

  Community norms should not impose costs on sharing such information. I’m sorry you had to pay these costs, Frances. Thank you for speaking out. Hopefully this post decreases the cost in these communities. In fact, such important information should be socially subsidized, not taxed (since e.g. speaking out often requires reliving trauma, which is unpleasant; and most of the benefit is external).

  What links here?

  • EA sex­ual ha­rass­ment: Where’s the ev­i­dence we’re worse than baseline? by Nathan Young (21 May 2026 16:32 UTC; 70 points)
  • Alexander Turner 🔸's comment on EA sex­ual ha­rass­ment: Where’s the ev­i­dence we’re worse than baseline? by Nathan Young (28 Jul 2026 17:46 UTC; -1 points)

An Opinionated Guide to Pri­vacy De­spite Authoritarianism

• Alexander Turner 🔸 22 Apr 2024 20:45 UTC

  8 points

  2 ∶ 0

  in reply to: Paul_Christiano’s comment on: De­cep­tive Align­ment is <1% Likely by Default

  Paul, I think deceptive alignment (or other spontaneous, stable-across-situations goal pursuit) after just pretraining is very unlikely. I am happy to take bets if you’re interested. If so, email me (alex@turntrout.com), since I don’t check this very much.

  I think that “deceptively aligned during pre-training” is closer to e.g. Eliezer’s historical views.

  I agree, and the actual published arguments for deceptive alignment I’ve seen don’t depend on any difference between pretraining and finetuning, so they can’t only apply to one. (People have tried to claim to me, unsurprisingly, that the arguments haven’t historically focused on pretraining.)