The dilemma at 40:30 is perhaps why we need agents. Since GPT-4 is the source LLM Agent 1 would attempt to get the answer from GPT-4 and if it cannot get the answer from your custom PDF corpus
Thank you for the video. I believe there's a key point missing though. While ChatGPT-4 excels in general knowledge, the strength of a PDF-focused tool lies in its ability to provide specific insights, like interpreting contract details or product requirements. It'd be interesting to compare how well these tools extract and summarize complex points scattered within a document, not just single references found via an index search.
For some one who knows how to code with openai API, this is really easy to do. I suspect there will be dozens of such websites very quickly, but not sure how long will the free access last because each PDF consumes a lot of tokens.
So whch one was the best at READING AND UNDERSTANDING A PDF ? not who answered questions correct about the topic, what was important is the ability to read and comprehend a pdf.
Hello, I just wanted to say that I really enjoy your content on RUclips. However, I've noticed that in some of your videos, your head can sometimes block important parts of the content. It can be a bit distracting and take away from the overall viewing experience. I was wondering if you could perhaps adjust the camera angle or position to avoid this in future videos? I think this small change could really enhance the quality of your videos. Thank you for considering my feedback.
I have tried ChatPDF and GPT4 and for most subjects GPT4 is the way to go, but for other ChatPDF does it better. Also for research you would need it to read pdf files etc.
@@hwoccurrence ChatPDF, is just langchain that takes a PDF, turns it into a vectorstore database, and queries it with the input you ask about the PDF. It could be used with any LLM, the 'chatPDF' website most likely uses GPT3.5, but its literally a single line of code they would change to use gpt4 instead. They just dont because the costs of gpt4 is 100x+ of gpt3.5 right now. And terms of 'looking over a pdf' and answering questions about it, GPT4 would be overkill as it does not perform that much better than GPT3.5. GPT4 key features are its context size(like 5x of gpt3.5),better knowledge of hard topics like law etc and the image inputs. But as for the general task of 'read this paragraph and tell me about x/y' its on par.
What is the vision and mission of this chat pdf and pdf-gpt and chat gpt 4 actually and is this a combination of google and youtube and language translation and designer logos and fashion and fashion graphics and stock trading and gamers and what's up and online shop and the latest second generation of intelligence.....?
ChatPDF and PDF-GPT were sometimes able to find the relevant information from the PDF files, but they also had some errors and limitations. ChatPDF was more succinct and did not error out, while PDF-GPT had a better UI and more references. But neither of them could match GPT-4's general knowledge and accuracy. So I would say GPT-4 is the best platform for answering questions about any topic.
ChatPDF is an interface, not a LLM. You can use chatPDF, or tech 'like it', with any LLM. Aka, you could use it with GPT4, but that is just stupid because GPT4 isnt better at GPT3.5 at many things besides amount of knowledge stored correctly and context size.
@@draken5379 ChatPDF and PDF-GPT are useful tools for answering questions from PDF files but they are not perfect. ChatPDF gives shorter and clearer answers but its interface is very simple. PDF-GPT offers a nicer interface but sometimes it gives errors or fills with unnecessary information. But the real problem is that these tools cannot reach the level of GPT-4. GPT-4 is much superior in general knowledge, accuracy and context understanding. Using ChatPDF with GPT-4 would not be stupid, on the contrary, it would be very smart.
@@wild7claw Once again, ChatPDF, IS chatGPT. ChatPDF, and PDF-GPT, are not 'models' or 'AIs'. They are interfaces. They USE GPT3.5, or 4 if they want to, they can set which model to use. They use a framework called langchain, to convert the pdf into a vectorstore database, query it, and it pastes that text into the conversation between you and GPT. GPT4 is only better in knowledge and accuracy, its not better at reasoning or context understanding, but it has a much larger context window, which HELPS with reasoning. Imagine it like the size of the AIs short term memory. Using chatPDF with GPT4 wouldnt be worth it. How chatPDF works, is its grabs the highest ranking paragraph from the vectorstore of the PDF, and places that in the conversation. These vectorstores have a max size per chunk. Using GPT4 will always be 'better', but that has nothing to do with any of this. For the task of, 'read this paragraph i just gave you from this pdf', they are very similar. But GPT4 is 100x the cost. It should only be used when it makes sense for the task at hand.
How about a very recent paper….something within the last month or so and/or something not on the internet as believe all these papers are part of the Arxiv dataset that is in the pre training for gpt4 so it’s not really apples to apples here….like the sparks of AGI paper just released…not sure gpt4 would know anything but if you add he papers to the other two, it would be able to discuss and Q&A over it….just pointing that out…..great comparison though🥳🦾🤓 1:14:05
The dilemma at 40:30 is perhaps why we need agents. Since GPT-4 is the source LLM Agent 1 would attempt to get the answer from GPT-4 and if it cannot get the answer from your custom PDF corpus
Thank you for the video. I believe there's a key point missing though. While ChatGPT-4 excels in general knowledge, the strength of a PDF-focused tool lies in its ability to provide specific insights, like interpreting contract details or product requirements. It'd be interesting to compare how well these tools extract and summarize complex points scattered within a document, not just single references found via an index search.
For some one who knows how to code with openai API, this is really easy to do. I suspect there will be dozens of such websites very quickly, but not sure how long will the free access last because each PDF consumes a lot of tokens.
ChatPDF is an interface, not a model or LLM.
All its doing, is opening up the PDF, and dropping snippets it from it, into calls to openAIs LLMs.
So whch one was the best at READING AND UNDERSTANDING A PDF ? not who answered questions correct about the topic, what was important is the ability to read and comprehend a pdf.
Is there another way of measuring understanding that doesn't consist of answering questions?
@@hu-po I don't think so...
I can confirm the volume is good
Friend, Does this read documents and summary for us, do you have any website reading pdf? any considerations?
Is it possible to delete folders and files from ChatPDF?
Hello, I just wanted to say that I really enjoy your content on RUclips. However, I've noticed that in some of your videos, your head can sometimes block important parts of the content. It can be a bit distracting and take away from the overall viewing experience. I was wondering if you could perhaps adjust the camera angle or position to avoid this in future videos? I think this small change could really enhance the quality of your videos. Thank you for considering my feedback.
This is good feedback, thank you.
Chatpdf couldn't read any of my pdf files after multiple attempts. Provided a lot of absolutely unrelated answers. I didn't understand why ...
I have tried ChatPDF and GPT4 and for most subjects GPT4 is the way to go, but for other ChatPDF does it better. Also for research you would need it to read pdf files etc.
ChatPDF, isnt an LLM, its an interface.
You can use chatPDF, with GPT4.
@@draken5379 How?
@@hwoccurrence ChatPDF, is just langchain that takes a PDF, turns it into a vectorstore database, and queries it with the input you ask about the PDF.
It could be used with any LLM, the 'chatPDF' website most likely uses GPT3.5, but its literally a single line of code they would change to use gpt4 instead.
They just dont because the costs of gpt4 is 100x+ of gpt3.5 right now.
And terms of 'looking over a pdf' and answering questions about it, GPT4 would be overkill as it does not perform that much better than GPT3.5.
GPT4 key features are its context size(like 5x of gpt3.5),better knowledge of hard topics like law etc and the image inputs.
But as for the general task of 'read this paragraph and tell me about x/y' its on par.
So who is the winner?
ChatGPT! It was able to answer questions corrrectly without even having access to the paper.
@@hu-po lol
How could I setup my camera like you ,I mean with pdfs on background😢
I use OBS studio
Out of topic, but how do his eyes get in front of his glasses? In one second I see glasses' frame, in the next second his eye pops up front 😅
I'm secretly an alien
@@hu-po :)) , But seriously is computer making some adjustments or is my brain playing tricks on me?
Lookup NVIDIA Broadcast
What is the vision and mission of this chat pdf and pdf-gpt and chat gpt 4 actually and is this a combination of google and youtube and language translation and designer logos and fashion and fashion graphics and stock trading and gamers and what's up and online shop and the latest second generation of intelligence.....?
ChatPDF and PDF-GPT were sometimes able to find the relevant information from the PDF files, but they also had some errors and limitations. ChatPDF was more succinct and did not error out, while PDF-GPT had a better UI and more references. But neither of them could match GPT-4's general knowledge and accuracy. So I would say GPT-4 is the best platform for answering questions about any topic.
ChatPDF is an interface, not a LLM.
You can use chatPDF, or tech 'like it', with any LLM. Aka, you could use it with GPT4, but that is just stupid because GPT4 isnt better at GPT3.5 at many things besides amount of knowledge stored correctly and context size.
@@draken5379 ChatPDF and PDF-GPT are useful tools for answering questions from PDF files but they are not perfect. ChatPDF gives shorter and clearer answers but its interface is very simple. PDF-GPT offers a nicer interface but sometimes it gives errors or fills with unnecessary information. But the real problem is that these tools cannot reach the level of GPT-4. GPT-4 is much superior in general knowledge, accuracy and context understanding. Using ChatPDF with GPT-4 would not be stupid, on the contrary, it would be very smart.
@@wild7claw Once again, ChatPDF, IS chatGPT.
ChatPDF, and PDF-GPT, are not 'models' or 'AIs'. They are interfaces. They USE GPT3.5, or 4 if they want to, they can set which model to use.
They use a framework called langchain, to convert the pdf into a vectorstore database, query it, and it pastes that text into the conversation between you and GPT.
GPT4 is only better in knowledge and accuracy, its not better at reasoning or context understanding, but it has a much larger context window, which HELPS with reasoning. Imagine it like the size of the AIs short term memory.
Using chatPDF with GPT4 wouldnt be worth it. How chatPDF works, is its grabs the highest ranking paragraph from the vectorstore of the PDF, and places that in the conversation. These vectorstores have a max size per chunk.
Using GPT4 will always be 'better', but that has nothing to do with any of this. For the task of, 'read this paragraph i just gave you from this pdf', they are very similar.
But GPT4 is 100x the cost. It should only be used when it makes sense for the task at hand.
Testing the comments are working
Testing. Testing
testiiiiing
Да, он не изучает весь документ а только маленькую часть и остальное придумывает. Анализ большого документа он не сможет сделать
🎉
How about a very recent paper….something within the last month or so and/or something not on the internet as believe all these papers are part of the Arxiv dataset that is in the pre training for gpt4 so it’s not really apples to apples here….like the sparks of AGI paper just released…not sure gpt4 would know anything but if you add he papers to the other two, it would be able to discuss and Q&A over it….just pointing that out…..great comparison though🥳🦾🤓 1:14:05