Blog

A one-dimensional PCA projection of OpenAI’s text-embedding-ada-002 model achieves 73.7% accuracy on the ETHICS Util test dataset. This is comparable with the 74.6% accuracy of BERT-large finetuned on the entire ETHICS Util training dataset. This demonstrates language models develop implicit representations of human utility purely from self-supervised learning.

Our adversarial testing algorithm uncovers a simple, human-interpretable strategy that consistently beats superhuman Go AIs. We explore the implications this has for the robustness and safety of AI systems.

All contemporary machine learning systems are vulnerable to adversarial attack. This poses serious problems for existing alignment proposals. We explore these issues and propose several research directions FAR is pursuing to overcome this challenge.