neverix

• neverix 14 Feb 2023 18:46 UTC

  2 points

  0

  in reply to: Quintin Pope’s comment on: Is In­struc­tGPT Fol­low­ing In­struc­tions in Other Lan­guages Sur­pris­ing?

  Is RLHF updating abstract circuits an established fact? Why would it suffer from mode collapse in that case?

• neverix 6 Feb 2023 15:57 UTC

  7 points

  0

  in reply to: Jessica Rumbelow’s comment on: SolidGoldMag­ikarp (plus, prompt gen­er­a­tion)

  It is based on this. I changed it to optimize using softmax instead of straight-through estimation and added regularization for the embedded tokens.

  Notebook link—this is a version that mimics this post instead of optimizing a single neuron as in the original.

  EDIT: github link

• neverix 5 Feb 2023 19:31 UTC

  3 points

  0

  in reply to: mic’s comment on: SolidGoldMag­ikarp (plus, prompt gen­er­a­tion)

  This project tried this.

• neverix 5 Feb 2023 19:26 UTC

  7 points

  0

  on: SolidGoldMag­ikarp (plus, prompt gen­er­a­tion)

  I did some similar experiments two months ago, and with to your setup the special tokens show up on the first attempt:

  

• neverix 26 Jan 2023 7:51 UTC

  1 point

  0

  on: Thoughts on the im­pact of RLHF research

  techn ical

  typo?

• neverix 30 Oct 2022 12:50 UTC

  1 point

  on: A His­tory of Bayes’ Theorem

  Our of curiosity

  typo

• neverix 4 Nov 2021 12:15 UTC

  4 points

  in reply to: 1a3orn’s comment on: What’s the differ­ence be­tween newer Atari-play­ing AI and the older Deep­mind one (from 2014)?

  s/​EfficientNet/​EfficientZero?