severe cove
We use GPT4 and some times ChatGPT to assist in complex text analysis.
Our use case include form generation, so we need markers and placeholders.
We feed GPT with the text (usually it is 2 A4 pages) and ask it to suggest how to code marked areas and what short placeholders can be used.
We definitely will switch to "private GPT" later, when we will have enough data to trane it on.
We are also interested on OCR2description when we can feed the image and get back what it is with some probability (say this is an invoice from company A, or a payslip etc). Main issue is that GPT4 is very restricted, so if there are personal details it most likely will reject the image, even if it is not asked to extract any.