Interesting snippet from the introduction:
Breakthrough capabilities – We demonstrate breakthrough capabilities in language understanding and generation across a number of difficult tasks. Specifically, Section 6.3 presents evaluation on a collection of reasoning tasks, which require multi-step mathematical or common sense reasoning to produce the correct answer. Prior state-of-the-art results use a combination of task-specific fine tuning, domain-specific architectures, and task-specific verifiers to achieve strong results. In this work, we demonstrate that when model scaling is combined with chain-of-thought prompting (Wei et al., 2022b), simple few-shot evaluation can outperform or match the fine tuned state of the art on a wide array of reasoning tasks. In Section 6.2 we additionally highlight breakthrough performance on BIG-bench (BIG- bench collaboration, 2021), a recently released suite of 150+ new language understanding and generation tasks, many of which are extremely difficult even for humans to correctly answer. In Figure 1 and Section 9, we demonstrate exploratory capabilities in PaLM’s ability to explicitly interpret and explain complex reasoning chains.