ChatGPT Powered Question Answering Over Documents
通过
Skanda Vivek |
已更新 a month ago |
Artificial Intelligence/Machine Learning
Health Check
N/A
返回所有讨论
Whole document is not being parsed in the context
pratikbharsakle
a year ago
I am sending a text from a PDF document. The text is approximately 50000 chars long and PDF is of approx 30 pages. I observed in the response, the content only shows part of text I sent as part of request. And the API is not able to answer the question whose answer is present on page 1 or page 2 of text submitted.
Would appreciate some help on this matter.
加入讨论 - 在下面添加评论：
登录/注册以发布新的评论
Hi - thanks for your question. Good observation that the response contains only part of the text. This is because ChatGPT has limitations on the number of max. tokens it can accept. Under the hood, the API is doing an embeddings comparison to find which chunks of text most likely contain the answer (which is why the reponse only has chunks of text).
Here is an article I wrote if you want more details - feel free to let me know if you have any other questions.
https://pub.towardsai.net/chatgpt-api-101-a-beginners-guide-95e6d41c716f
Best,
Skanda