JanB

• JanBJul 29, 2021, 5:59 PM

  33 points

  1

  on: How much do vari­a­tions in diet qual­ity de­ter­mine in­di­vi­d­ual pro­duc­tivity?

  Hi,

  I’m the PI (supervisor) on the replication of the creatine trial. We have finished the trial but not the data entry+analysis. Extremely tentatively, it looks like the effect replicates but with much smaller effect size. If this result remains after data entry+analysis are complete, then I’d say this is a positive update for cognitive benefits of creatine—given that the prior for this to work was pretty low (publication bias, and so on), I’d say that one trial with a large effect size and one with a small effect size is more convincing than just one with a large effect size.

  But really, the result might still change in the process of us double-checking things. It’s just my best guess right now. The preprint will hopefully appear in the course of the year.

• JanBFeb 16, 2020, 5:39 PM

  LW: 4 AF: 2

  AF

  in reply to: evhub’s comment on: What is the differ­ence be­tween ro­bust­ness and in­ner al­ign­ment?

  Great answer, thanks!

• JanBFeb 16, 2020, 5:35 PM

  LW: 2 AF: 1

  AF

  on: Does iter­ated am­plifi­ca­tion tackle the in­ner al­ign­ment prob­lem?

  Thanks for all your answers :-)

[Question] What is the differ­ence be­tween ro­bust­ness and in­ner al­ign­ment?

[Question] Does iter­ated am­plifi­ca­tion tackle the in­ner al­ign­ment prob­lem?

• JanBFeb 7, 2020, 2:21 PM

  LW: 1 AF: 1

  AF

  on: Disen­tan­gling ar­gu­ments for the im­por­tance of AI safety

  I struggle to understand the difference between #2 and #3. The prosaic AI alignment problem only exists because we don’t know how to make an agent that tries to do what we want it to do. Would you say that #3 is a concrete scenario for how #2 could lead to a catastrophe?

The ex­pected value of ex­tinc­tion risk re­duc­tion is positive