EmpressCelestia

• EmpressCelestia 24 Mar 2025 1:16 UTC

  7 points

  1

  in reply to: Raemon’s comment on: Eliz­a­beth’s Shortform

  I feel like it should just loop back around to being cool at that point but I guess that’s not how it works.

• EmpressCelestia 25 May 2022 3:43 UTC

  10 points

  0

  on: A Quick Note on AI Scal­ing Asymptotes

  This paper (https://​​arxiv.org/​​abs/​​2010.14701) shows the existence of constant terms in other generative modelling settings and relates it to the entropy of the dataset, where you can’t compress beyond. It also gives empirical evidence that downstream performance in things like “finetuning a generative model to be a classifier” continues to improve as you asymptote to the constant. From a physics perspective, the constant term and coefficients on the power law pieces are “non universal data” while the exponent is going to tell you more about the model, training scheme, problem, etc.

• EmpressCelestia 2 May 2018 0:00 UTC

  1 point

  0

  in reply to: mwengler’s comment on: Wel­come to San Diego Ra­tion­al­ist Mee­tups

  The Google group that will give you access to event announcements and the calendar is here: https://​​groups.google.com/​​forum/​​#!forum/​​san-diego-ratadj

  There’s also a FB group: https://​​www.facebook.com/​​groups/​​1944288489124568/​​