Subscribe
Sign in
Not to rain on anybody's parade, but this product has a VERY small moat. Here's a free version of it that requires using your own OpenAI key: https://knowledgegpt.streamlit.app/ (I'm not affiliated with this product in any way -- I just happened to be curious about Streamlit.) The PDF.ai design is really nice. It's very well executed. I just doubt there's much of a long-term future here.
I used this product, and it is fast. But it gives wrong results. That is a problem as it can mislead the user. Maybe because it uses a cheaper embedding to save cost? I think this needs to be fixed before launching it widely. Lifetime pricing is nice but the product should work properly. It is not disclosed if this uses Open AI and which version. I wish I could rate it better but it needs improvement before you sell it. All the best, Amir
I purchased this LTD a few days back when it was promoted as a limited-time offer that will "go away in no time." Kudos to the team. Scarcity marketing works wonders for AI apps these days, and it seems the trick is also working for their PH launch. I noticed some discussions about app performance, which I believe is a genuine concern. I created a Twitter thread to provide feedback a few days ago, but it went unnoticed. Please improve the app since it's not working well. 🤞
I uploaded a book and it gave me: "I'm sorry, I cannot summarize the 10 key points as the document(s) provided are not specified. Please provide more information or context so I can assist you better." I gave it the title and asked if it had it indexed. Here is the response: "As an AI language model, I do not have access to your uploaded document or any external database. I apologize for the inconvenience. Is there anything else I can assist you with?" I then asked for the title and it said: "I'm sorry, but based on the extracted parts of the document provided, I cannot determine the title of the document. The parts provided only show that there are several pages intentionally left blank. Please provide more information or context so I can assist you better." I asked for the page numbers of the blank pages. It gave those to me which was perplexing since I had assumed up until this point that the document did not load. I then told it that the title of the document was a book title on 5 different pages, which included the first 2 pages. It then confirmed that it could find the title. I asked why it did not find it initially and it gave the following response: "I apologize for any confusion. I may have misunderstood your initial question. If the document you were referring to is a book titled "XYZ" then I can confirm that I do not have it indexed or stored in my database. As an AI language model, I do not have access to external databases or documents unless they are provided to me by the use". So it's clear that the pinecone or other vector database code missed big time. By the way, this product idea is out there in spades in the open source arena. Not sure it's worth $99 for the Google Collab and Pinecone costs. Let me know your thoughts.
There is a reason for the "special" lifetime pricing that was first announced as a very limited, soon-to-disappear deal by Damon. He has now made it a perpetual pricing model to gather as many signups as possible and launched on PH with the same intent. Nothing sells like scarcity especially when AI is attached to it. The app is broken, and the results are inaccurate and useless. I bet anyone who is upvoting has not used the app and is joining the cheering crowd. PDF.ai is a disaster and not even worth calling an alpha. Buying someone else's work and wrapping it under a fancy domain is all that Damon has done.
Sadly this app doe not work - after wasting an hour or so and using four different PDFs the app did not work. Error -= Failed to Fetch Document No stars but there is no such option - so One star for me
I purchase the lifetime subscription and I'm not happy. I've been struggling for quite some time to get some help with a few documents and the responses have been mostly useless. I've tried to simplify my requests to the most simple questions I can think of, such as how many questions are on a page and it doesn't seam to even understand what a question is. This image should demonstrate how frustrating this is: https://prnt.sc/Mqw7epWzP6Us.
I paid for the ability to chat all pdfs. It can't even do something basic like list the titles of less than 10 pdfs uploaded, but will tell you the title if you chat one at a time. Pretty hard to have faith in what it can do on that alone. I've had zero help response and noticed youtube media strongly promoting it's chat all function which in my experience does not work at all. I can't emphasise strongly enough to avoid this product sadly.
I just tested the demo and it works really well, as long as you formulate a question formally. Here's one thing that did not work for me and I will use this as a benchmark to see the quality of the product improves. "Can you extract any emails from the document?" This was wrong, the email was located on page 1. Same with author's name.
This is a great idea - so many PDF chatbots; this is very much a cool space. Can you share what your vector database is helping you do? You mention it, but you don't say why that's special. Maybe that will be something that helps your marketing!
Quicky AI