While checking the capability of large language models - I could see that the capability is good enough to extract required values from structured and unstructured documents
With above details - What is the future of document understanding module where we need to train the documents by labelling
In my opinion, LLM is more of general purpose. It means, it can do everything possible with that LLM model.
Where as Document understanding uses Specialized AI which is targeted to the specific problem, Understanding various format and types of documents.
Document understanding modules provides you specialized tools to automated documents whereas LLM models are large scope which can be used for document automation but will require relatively more efforts than DU.
If you see modern DU the base model is an docpath LLM…which is specifically designed for DU…
So going forward even models from AI center are being deprecated as announced and replacement are DU , Gen AI and Comm path all using different LLM specialized for tasks so yes LLM is being used in all
its not the DU would be replaced but the backend models used will definitely be better…so even in DU modern now there is predictive labellign as well which helps in labelling which automatically identfies the labels as well
Yes autopilot is doing well too…but for specific large documents a specialized LLM can perform better as we know about the halicinations from LLM over time