Published onAugust 29, 2021Evaluating AI21 Jurassic-1 Jumbo with GPT-3 Davinci using Water Cooler Trivia datasetThis post evaluates AI21's Jurassic-1 Jumbo language model and compares it to OpenAI's GPT-3 davinci model in answering trivia questions from the Water Cooler Trivia dataset. The experiment was conducted in both zero-shot and prompt settings, with the former resulting in a 55.4% accuracy rate for Jurassic-1 and the latter resulting in a 46.15% accuracy rate.Read more →