Ashwin Kalyan

TL;DR: This work designs language models to learn to generate lectures and explanations as the chain of thought (CoT) to mimic the multi-hop reasoning process when answering S CIENCE QA questions and explores the upper bound of GPT-3 and shows that CoT helps language models learn from fewer data.

TL;DR: The Neural Guided Deductive Search (NGDS) as discussed by the authors is a hybrid synthesis technique that combines the best of both symbolic logic techniques and statistical models to synthesize user-intended programs from a small number of input-output examples.

TL;DR: The authors proposed Neural Guided Deductive Search (NGDS), a hybrid synthesis technique that combines the best of both symbolic logic techniques and statistical models to produce programs that satisfy the provided specifications by construction and generalize well on unseen examples.

TL;DR: The LILA Mathematical Reasoning Benchmark (LILA) as discussed by the authors is a general-purpose mathematical reasoning benchmark consisting of 23 diverset tasks along four dimensions: mathematical abilities e.g., arithmetic, calculus, language format e. g., question-answering, fill-in-the-blanks (iii) language diversity eg., no language, simple language, external knowledge eg. commonsense, physics).

TL;DR: A novel approach is proposed, P ROMPT PG, which utilizes policy gradient to learn to select in-context examples from a small amount of training data and then constructs the corresponding prompt for the test example, which outperforms the best baseline on the accuracy metric and reduces the prediction variance.