PlayfulTechnology/QARAC · Apply for community grant: Personal project (gpu and storage)

QARAC (Question Answering, Reasoning and Consistency) is a research project aimed at improving the factual accuracy of NLP models. The project aims to train a set of models that transform sentences to and from a vector space in such a way that logical relationships between the sentences map to arithmetic relationships in the vector space.
It should then be possible to store a knowledge base of trusted facts in a vector database, and use the models to query it, decode the answers, and possible infer answers from existing knowledge.

For more information, see https://playfultechnology.co.uk/qarac-question-answering-reasoning-and-consistency.html