Useful and practical ai and automations info you can use today
“And if it isn’t easy for ME, then it won’t be easy for YOU! I want to be able to upload a document, or point to a folder of documents, and have the AI read them, categorize them, and allow me to ask questions about them. My results to date have been very frustrating.”
And I don’t mean “common” in any negative way!
For me, AI needs to be at an everyday level — something you and I and my mom can understand and work with. Ok, my mom is 87 and struggles with cell phones and remotes, so maybe not her.
ChatGPT is easy. It is chat. Ask it a question and get a result. There is no trickery, no complicated anything to get a usable result.
Most other tools so far strike me as either being the same thing, or being more complicated.
And if you have looked under the covers at what a chatbot engine looks like, then you know it gets into VERY technical waters very quickly.
This week I started playing with Claude 2.0 and assumed it would be another ChatGPT knockoff. Boy was I wrong!
This is just me talking about my experiences here. I’m not an affiliate for them or anything…just a guy playing with a chatbot like the rest of you.
My angle is that I am building software that is a copilot for business people. It uses AI to help you get your job done. And instead of working from a blank canvas, I provide useful prompts, workflows, data sets, and dashboards that make sense for your role.
So the thing I am looking at these days is how well chat engines read, categorize and search PDFs and other documents that we all have strewn across our file systems.
It is surprisingly difficult to get it right! It sounded easy at the start, but the more I try, the more I see that these chat engines are good with some things but not others. (See the Model definition in the Learning Cube below for more info)
And if it isn’t easy for ME, then it won’t be easy for YOU! I want to be able to upload a document, or point to a folder of documents, and have the AI read them, categorize them, and allow me to ask questions about them. My results to date have been very frustrating.
I have been testing with a difficult document from the start. A PDF with lots of images and text. A guy on LinkedIn, Brian Julius, is a pro data analyst and he went up against ChatGPT’s Code Interpreter (see the last issue of the Practical AI newsletter for more on Code Interpreter and Brian’s post). He created a carousel of the results, which I downloaded to PDF.
(You can download any carousel as a PDF, did you know that?)
I then went to https://claude.ai and created a free account. It gave me a chat box with a file upload button on it. Bam, grabbed the PDF and uploaded it. Then I simply asked “what is this document about?”
The results came streaming back almost immediately. Impressive!
But more impressive was the wording and accuracy of the results. As you can see, they are very “chattish” and not as robotic as ChatGPT is with its generic results.
The next document I tried was a very biologically technical document I pulled from PubMed. My brother is a neurobiologist and reads these research papers to see how certain proteins are being expressed.
They are NOT common everyday English. And they are VERY technical. Terms galore. I’ve had mixed results working with them in ChatGPT and other tools.
Gave it to Claude and results were pretty amazing!
To really grasp how cool this is, consider that the first document is plain old English but wrapped around images. So it is pretty much just looking at English text and determining meaning.
The same AI then takes a super geeky, jargon-filled research paper that is in a line-numbered lots of white space format, with charts and tables throughout…
…and handles it just fine. I also put both documents into the same chat thread, thinking it might throw Claude off a bit. Nope.
Anyway, the tools are getting better and I see some great use cases for something like Claude in your business:
Take old PDFs that have poor naming and throw them into Claude to create an executive summary of the document that you can then store with it (or throw away as no longer needed)…all without having to open the document, scan through it, determine what it is saying, etc etc. 5 seconds and you can see the totality of it and decide to keep or throw.
Review a newsletter you are writing and rewrite it in your “voice”
Feed it like 10 samples of your writings and tell it that is your writing style.
Then feed it your draft and say to rewrite it to match your style plus make it easy to read, etc etc.
I know this is common in ChatGPT too but the results with Claude are a step above anything else from what I can tell
Translate your user guides into Spanish and other languages
Ask it to help you with Focus Batching
Enjoy this content? Sign up for my weekly newsletter, "Practical AI", where I cull through all the noise to bring you the practical stuff from the AI field that you can do today in your business.
gimmee.info 2023 - All Rights Reserved - Terms & Conditions