Tomek Korbak
@tomekkorbak
가입 June 2017
644 팔로잉 중 11K 팬
understanding generalization is key to understanding misalignment. automating that understanding could be a powerful way to accelerate alignment research. we take a stab at this by using AIs to predict the alignment effects of training just by looking at the training data
더 보기
Can you tell whether an AI will become evil by looking at its training data?
We built an AI forecaster that reads training data and predicts whether a model will exhibit various forms of misalignment, such as deception or power-seeking, before training.
It performs well above random guessing and beats several much stronger baselines 🧵
Paper:
더 보기
