WebGPT: Improving the factual accuracy of language models through web browsing

OpenAI Blog · 4 years ago

OpenAI fine-tuned GPT-3 to answer open-ended questions more accurately by giving it access to a text-based web browser. The model improved from 68% to 79% accuracy on a factual QA benchmark when equipped with browsing capability. This allows language models to retrieve current information and verify facts rather than relying solely on training data.