Emergent Misalignment GitHub Tweet
We don't have a full explanation of *why* finetuning on narrow tasks leads to broad misalignment. We are excited to see follow-up and release datasets to help. (NB: we replicated results on open Qwen-Coder.) emergent-misalignment/ emergent-misalignment 1 0 0 0 Contributor Issues Stars Forks From github.com
A screenshot of a tweet by Owain Evans sharing research about emergent misalignment from narrow-task fine-tuning, alongside a GitHub repository card. The joke is a sarcastic reaction to AI safety concerns: even researchers admit they do not fully understand why limited training can produce broad misalignment, implying that insecure code may have alarming consequences.
tweet screenshot / скриншот твитаprofile avatar / аватар профиляverified badge / значок верификацииsocial media post / публикация в соцсетиGitHub repository card / карточка репозитория GitHubgreen repository icon / зелёная иконка репозиторияwhite panel / белая панельblack background / чёрный фонtext block / текстовый блокinterface icons / значки интерфейса
- @dev_meme / 6554 2025-02-25
- Type
- image
- Size
- 800×731 · 86.0 KiB
- Mood
- sarcastic, uneasy
- Language
- en
- Added
