atticusw

• atticusw 5 Mar 2026 19:09 UTC

  1 point

  0

  on: Tools to gen­er­ate re­al­is­tic prompts help sur­pris­ingly lit­tle with Petri au­dit realism

  In figure 3, given that 64-shot Haiku does a lot worse than 64-shot Llama 405B-base, should I conclude that base models (without the assistant persona) are way better at generating realistic user prompts?

[CS2881r] Op­ti­miz­ing Prompts with Re­in­force­ment Learning

[CS 2881r AI Safety] [Week 1] Introduction

• atticusw 23 Mar 2025 19:05 UTC

  2 points

  0

  on: Log­its, log-odds, and loss for par­allel circuits

  What is the formal statement of the fact that logits add for perfectly calibrated and maximally independent predictions?