WebGPT: A Next-Generation Models That Can Using Search Results
WebGPT was developed to improve the answer accuracy of the GPT-3 model. Due to the fixed text format of GPT-3’s output, many open-ended questions result in inaccurate answers and it is difficult to determine their source, making it challenging to verify their accuracy. To enhance the factual accuracy of the answers, OpenAI fine-tuned the GPT-3 model to create the WebGPT model.
What is WebGPT?
WebGPT is the new generation model developed by OpenAI based on the fine-tuning of GPT-3, which can use text information in a browser to respond to open-ended questions. With WebGPT, answer sources can be displayed, effectively enhancing answer accuracy. Its process is consistent with human online query answering – submitting a search query, tracking links, and scrolling up and down web pages. OpenAI announced information about WebGPT on its website on December 16, 2021.
After training, WebGPT can answer questions from ELI5, a dataset of open-ended questions collected from the “Explain Like I’m 5” subreddit.In its announcement, OpenAI revealed that “ We trained three different models, corresponding to three different inference-time compute budgets. Our best-performing model produces answers that are preferred 56% of the time to answers written by our human demonstrators, with a similar level of factual accuracy. Even though these were the same kind of demonstrations used to train the model, we were able to outperform them by using human feedback to improve the model’s answers.”
The Important Information of WebGPT
- On December 16, 2021, OpenAI announced the release of WebGPT on its website
- According to media reports, a new version of WebGPT may be released at some point in 2023