7 Comments

Seems like an overkill for this specific usecase. Free tools such as tesseract have OCR and do a fairly good job.

Expand full comment

@Unmesh, for most people, this is a big improvement over special purpose tools because:

1. No need to download a specific tool; (same reason why phone cameras have killed real cameras)

2. No need to learn a new tool

3. And most importantly, the ability to do far more than simple OCR and table extraction. You can't get tesseract to do anything that it wasn't actually programmed to do; But in case of ChatGPT, doing that is just a matter of tweaking the prompt. For example, I just did this: "Create a table containing a list of passengers, their gender, and their seat number from this eticket". And it did a perfect job, in spite of the fact that 3.1) the PDF did not actually contain a table, the data had to be extracted from different sentences, 3.2) the gender had to be inferred from the name prefix.

I'll try to give a few more example in future posts.

Expand full comment

Interesting - we use tesseract to provide data extraction facility as an enterprise offering. Will ask the team to explore this. Thank you.

Expand full comment

can we upload PDF to chatgpt for getting data ? if yes how ?

Expand full comment

Not in the default interface (at chat.openai.com). There you have to copy-paste the text.

But for those who know programming (or those who an use ChatGPT to get a program written) it would be quite simple to write a program to automate this.

Expand full comment

great , can you please send me the any reference tutorial for the same

Expand full comment

Will try to write something. In the meantime, check out https://chatpdf.com

Expand full comment