Tool Augmented Language Models. abs: arxiv.org/abs/2205.12255 Smaller tool augmented models outperform larger non-augmented models in two domains (thus far), and on out-of-distribution examples. Great collaboration w/AaronParisi and @YaoZhaoAI!

2+2=5? “LLMs are not Robust to Adversarial Arithmetic” a new paper from our team Google DeepMind with bucket of kets, Laura Culp, AaronParisi, Gamaleldin Elsayed, Jascha Sohl-Dickstein, Noah Fiedel TLDR: We ask an LLM to attack itself and find this works extremely well.