๐ก๐ผ๐ฏ๐ผ๐ฑ๐ ๐๐ฎ๐ธ๐ฒ๐ ๐๐ฝ ๐๐ต๐ถ๐ป๐ธ๐ถ๐ป๐ด: โ๐ ๐ป๐ฒ๐ฒ๐ฑ ๐๐ผ ๐ฐ๐ต๐ฎ๐ ๐๐ถ๐๐ต ๐๐ต๐ฎ๐ ๐ฃ๐๐!โ
Whatโs the point of โchatting with a PDFโ?
OpenAI now starts rolling out that use case where you can upload a document or a small set of documents (I heard of a limit of 5) to start asking it questions.
This may seem like a novel, new exciting feature, to many. But it has been around for ages and is the prototype example, the first โstarter codeโ (AKA โHello Worldโ) whenever you look into the setup for RAG (Retrieval Augmented Generation).
The concept is simple: slice up that document, store it in a database, and inject relevant parts back into the prompt.
Thereโs also been a couple software startups that have been doing that with varying success and functionality, generally known as โChat with PDFโ apps. These apps may come under pressure now more.
But my question is of another type:
If I have a specific document that I would like to review, then this is maybe a valid use case. Maybe I have this 80 page financial report, and didnโt find the paragraph headed โManagement Summaryโ, or maybe Iโm not able to find all the the details manually. Granted.
But I do NOT have just one specific document to analyze.
I have
โข product descriptions
โข marketing content
โข legal terms & conditions
โข prior customer communication
โข internal SOPS
just to name a few ๐จ๐ณ๐ฐ๐ถ๐ฑ๐ด of documents that want the AI to be aware of.
Thereโs no way Iโm going to find and upload the relevant documents to every prompt I write.
In fact, the pre selection to the relevant documents is like cheating, itโs like instructing Google to search only very specific parts of a website that I know well and that should have the relevant answer. It misses a huge part of the โsearch intelligenceโ.
I want to get help from AI to
โข write a sales response, and have relevant product features suggested from my manuals and website
โข respond to a customer complaint with all details about our terms & conditions, including links
โข write a new blog post that helps me internally link to other relevant blog posts on the topic
โข give a new employee tips and links to our SOPS, via a prompt where they ask questions - or another one where I can quickly summarize and reference relevant material
These are some use cases that we at AIPRM look into, and work on at the moment.
This will probably be the most powerful feature AIPRM has delivered so far, still in active development and test.
But some of our most active AIPRM Elite (and up) users will soon get an invite to try our upcoming feature โCustom Indexesโโฆand that will allow to accomplish such use cases.
I canโt wait!
๐๐ฏ๐ต๐ช๐ญ ๐ต๐ฉ๐ฆ๐ฏ ๐บ๐ฐ๐ถ ๐ค๐ข๐ฏ ๐ต๐ณ๐บ ๐ฅ๐ฐ ๐ต๐ฉ๐ฆ ๐ฎ๐ฆ๐ฏ๐ต๐ช๐ฐ๐ฏ๐ฆ๐ฅ ๐ธ๐ช๐ต๐ฉ ๐ต๐ฉ๐ฆ โ๐๐ฉ๐ข๐ต ๐ธ๐ช๐ต๐ฉ ๐๐๐โ ๐ง๐ฆ๐ข๐ต๐ถ๐ณ๐ฆ.