Humata is actually very good. I tried to converse with it using a legal case, and it was giving insightful answers. The only issue I had with it was that it freezes ever so often, then I would have to start over. But I usually ask the last question I had in mind, then it starts to pick the context of the conversation again
Dear Idea Cove, Yes, I agree. Coincidentally, I also uploaded a legal case and the analysis it gave was excellent. I actually spent a lot of time on it, but I didn't post any video since I am not a legal expert and did not feel comfortable opining about the law. But I did find the logic of the arguments provided to be quite sound and very well laid out. I actually found the answers to be very thoughtful in relation to the case. Best, Mike
Great video! My lab at Berkeley uses Humata AI because the answers are superior. Humata popular with the students here in UCB because it highlights references in the articles and saves responses, which is really helpful.
Wow. That is great. I'm not sure if it is "superior." They might be more scholarly maybe because of the citation and highlighting of the text, but ChatPDF is also able to explain the text, or that is, answer the queries with equal aplomb. I think it is just a tad more general or public facing in terms of its prose and writing voice. But both are useful for an undergraduate student researching the pdf... or just for a general audience user who wants to question the pdf instead of read it.
@@UniversityAIed We like the highlighting and Humata does a better job with summarization with cited references to the text. Their support and other tools is also super good.
@@UniversityAIed Humata lets us ask questions across multiple PDFs at the time. That's a big game changer and one of the reasons why it's superior to chatpdf. Humata also let's you see the PDF side-by-side and includes highlighting and references in addition to multi-file Q&A. That's why it's spreading across different labs and the entire student body, not just grad students in lab.
Excellent video! Quick question: The use case that you demonstrated (and most other demos typically go with) is feeding it a book/article and asking questions to summarize it. But what if the PDF I used was a contextualization of an API for a specific scripting language, with encapsulated methods/interfaces? If the PDF contained all possible methods and interfaces, complete with a tutorial and breakdown of example scripts that work correctly, and I asked it to generate a script to do X Y or Z (and these scripts aren't included exactly in the PDF itself but the methods to make them are), would either of these two programs be able to do so?
Dear Hussain, Great question. I have no idea! I think this means you will have to test it out with the very example you provided! I have a sinking suspicion that the authors of these interfaces wrote their code specifically for an interrogation of the uploaded text. You are suggesting a very different use model. You want to ask the interface to read the paper and then create a script. I'm not sure it is designed to do that. But you should give it a shot! And please report back here if you do try it. I would love to know if that works. Best, Mike
Sadly I don’t know a program that would do that. Hmmmm. … it is a good question; what a great feature if there was such a thing. But that would also kind of take the user out of the equation since there wouldn’t be any “chatting” that occurs between you and the documents.
@@thesocialjusticecentre3736 I want to automate this process. The problem with copying the questions from a PDF is formatting is out of whack. It would be nice for a program to recognize clearly formatted questions from a PDF.
They are both good. But unfortuntely, seems they both cannot search more than 1 file. I have to repeat my question 10 times if the answer locates in 10 files. Right?
Dear Ming Tak Wong, Yes, you are absolutely right. It allows you to interact and chat and make requests of only ONE document at a time. But that is also its strength because it allows the user to control where the question is directed. You can isolate ChatGPT's field of vision. But that is also the limitation because it cannot compare and contrast multiple documents. But one caveat: It will also look outside of the PDF document if you prompt it to look outside, and then it will draw on its wider training set of documents. However, then you cannot control what it looks at because you can only provide one PDF at a time. One solution could be to combine a bunch of articles into one longer PDF and then upload it.
Dear Dr. Irshad, Thank you for the find words. It is a very exciting and confusing time for the academy. Later on, I will also review another chat document interface called chatdoc: chatdoc.com Best, Mike
Dear Evelyn, Many thanks for the comment. I appreciate the kind words. I hope the comparison helps. Later on, I will also review another chat document interface called chatdoc: chatdoc.com Best, Mike
Dear Hamza, I have not tried it with epubs. But if you have epubs, then you can easily convert them to pdf with free online converters. I have used this one before: www.freeconvert.com/epub-to-pdf Best, Mike
Which limitations do you see for these types of interfaces regarding whether or not they contribute to humanities research work? Thanks for the video! :)
Dear Automata Learning Lab, Thanks for the question! re: limitations The key limitation is the chat format and the interface's chat limitation with ONE document. You can only interact with the document through questions or chat as the interface tries to makes it easier for the user to control the parameters of the discussion or the chat by limiting it to ONE document. But that is also an intellectual or conversational limit because it does not allow the user to compare an contrast the document with another. A real-life interlocutor might say something like "Oh, that makes me think of this other thing I was reading, and it says..." So, that is both the power and limit of interfaces that allow a user to keep ChatGPT in the "lane" of one specific document. re: contribution The ability to chat or talk to a specific document. is the power of ChatGPT because the format of asking "someone" to explain something is very natural and we are already primed to learn in this fashion. I think if you load up a PDF of Plato and Socrates --because of the Socaratic format of these dialogues-- the LLM transformer can almost make it feel like you are chatting with these long dead authors. In that sense, I can see how this can lead to the level of the uncanny in the future. Someone just needs to create some holographic technology and some speech to text interface. Although it could also get very repetitive because their words would be limited to a set vocabulary and pre-loaded content. There can never be a "new" set of thoughts or examples or new patterns of speech. But endless blending of existing speech and semantic parameters. But that is just idle specuation. re: contribute to humanities research Yes, sure. Programs like this already allow someone like me to pose new or unusual questions just to see what happens. And as we know, accidents are the mother of invention. These interfaces also creates joy or fun. So, making it fun to interact with an otherwise dry text, for example, might actually make it "more" compelling for a reader and lead to new ways of thinking about the text.
Dear Hamza, Yes, there are many interesting sites that can help someone with academic research and writing. Have you tried ChatDOC or Paperpal? Paperpal is like Grammerly. Best, Mike
Great video, thanks Mike. I don’t really understand how ChatPDF relates to ChatGPT. But if I upload a proprietary pdf to ChatPDF, will that pdf end up being part of ChatGPT’s data and essentially in the public domain? Much appreciated. Mark.
Dear Mark, ChatPDF is paying for use of the ChatGPT product. So, the authors of ChatPDF have written a program that allows for the user to chat with just one document. It is ChatGPT under the hood, but the interface is ChatPDF. re: will that pdf end up being part of ChatGPT’s data and essentially in the public domain? No, I do not think it will be part of public domain. But OpenAI is looking at everything you enter into its system, and the Large Language Model of ChatGPT is "looking" at your uploaded text; however, it is not storing that data and incorporating it into the training data it was exposed to during its training and fine tuning phases. Will it remember that data? I do not think so because it is not part of the training data of the original version (e.g. 3.5 or 4). I think it only remembers it during your engagement with it. And even then it often will lose the thread and go off on tangents that clearly demonstrate it has forgotten what you showed it earlier. But I could be wrong. Best, Mike
Dear Kendall, Yes, it works with other languages. I just uploaded a French document and it was able to answer my questions in English. So, just like ChatGPT, it is good with other languages. What language did you want to use? Best, Mike
Dear Simone, Thank you for the kind words. And thanks for the suggestion. I had not heard of it before. So, that is super useful. I will try it out and post my thoughts. Best, Mike
Dear Uijwal Chetan, Do you mean the co-pilot on Bing Chat? Or do you mean Github co-pilot? I don't know much about Github co-pilot but I think that is dedicated to writing code. There is an assistant in Bing that is kind of like a co-pilot that can help you with questions and prompts, but I do not think it allows for uploading a specific PDF. I do like Bing Chat as it runs on a version of Chat GPT 4, so the answers are quite good. Best, Mike
Dear Mila, I think HUMATA gives you a certain amount of usage for free every month. If you exceed it, then you can always use ChatDoc or ChatPDF. Best, Mike
Dear Cyberphunkisms, I am a prof that is afraid of AI! I'm afraid because I know how powerful a tutor and research tool it can be. I am afraid that it might be far easier for a student to ask the AI to do all their work for them instead of struggling to do it themselves. That is the real risk out there now. Best, Mike
Humata is actually very good. I tried to converse with it using a legal case, and it was giving insightful answers. The only issue I had with it was that it freezes ever so often, then I would have to start over. But I usually ask the last question I had in mind, then it starts to pick the context of the conversation again
Dear Idea Cove,
Yes, I agree. Coincidentally, I also uploaded a legal case and the analysis it gave was excellent. I actually spent a lot of time on it, but I didn't post any video since I am not a legal expert and did not feel comfortable opining about the law. But I did find the logic of the arguments provided to be quite sound and very well laid out. I actually found the answers to be very thoughtful in relation to the case.
Best,
Mike
Great video! My lab at Berkeley uses Humata AI because the answers are superior. Humata popular with the students here in UCB because it highlights references in the articles and saves responses, which is really helpful.
Wow. That is great. I'm not sure if it is "superior." They might be more scholarly maybe because of the citation and highlighting of the text, but ChatPDF is also able to explain the text, or that is, answer the queries with equal aplomb. I think it is just a tad more general or public facing in terms of its prose and writing voice. But both are useful for an undergraduate student researching the pdf... or just for a general audience user who wants to question the pdf instead of read it.
@@UniversityAIed We like the highlighting and Humata does a better job with summarization with cited references to the text. Their support and other tools is also super good.
Hook it up! What other AI apps you guys use?
@@UniversityAIed Humata lets us ask questions across multiple PDFs at the time. That's a big game changer and one of the reasons why it's superior to chatpdf. Humata also let's you see the PDF side-by-side and includes highlighting and references in addition to multi-file Q&A. That's why it's spreading across different labs and the entire student body, not just grad students in lab.
Have you tried Claude AI? Its an LLM that also allows you to interact with documents.
Excellent video! Quick question:
The use case that you demonstrated (and most other demos typically go with) is feeding it a book/article and asking questions to summarize it. But what if the PDF I used was a contextualization of an API for a specific scripting language, with encapsulated methods/interfaces? If the PDF contained all possible methods and interfaces, complete with a tutorial and breakdown of example scripts that work correctly, and I asked it to generate a script to do X Y or Z (and these scripts aren't included exactly in the PDF itself but the methods to make them are), would either of these two programs be able to do so?
Dear Hussain,
Great question. I have no idea! I think this means you will have to test it out with the very example you provided! I have a sinking suspicion that the authors of these interfaces wrote their code specifically for an interrogation of the uploaded text. You are suggesting a very different use model. You want to ask the interface to read the paper and then create a script. I'm not sure it is designed to do that. But you should give it a shot! And please report back here if you do try it. I would love to know if that works.
Best,
Mike
I need a program that can read one document to answer questions in another uploaded document. Any suggestions?
Sadly I don’t know a program that would do that. Hmmmm. … it is a good question; what a great feature if there was such a thing. But that would also kind of take the user out of the equation since there wouldn’t be any “chatting” that occurs between you and the documents.
Question: could you just copy paste each individual question? Or do you want to automate that process? How many questions are there?
@@thesocialjusticecentre3736 I want to automate this process. The problem with copying the questions from a PDF is formatting is out of whack. It would be nice for a program to recognize clearly formatted questions from a PDF.
They are both good. But unfortuntely, seems they both cannot search more than 1 file. I have to repeat my question 10 times if the answer locates in 10 files. Right?
Dear Ming Tak Wong,
Yes, you are absolutely right. It allows you to interact and chat and make requests of only ONE document at a time. But that is also its strength because it allows the user to control where the question is directed. You can isolate ChatGPT's field of vision. But that is also the limitation because it cannot compare and contrast multiple documents. But one caveat: It will also look outside of the PDF document if you prompt it to look outside, and then it will draw on its wider training set of documents. However, then you cannot control what it looks at because you can only provide one PDF at a time. One solution could be to combine a bunch of articles into one longer PDF and then upload it.
Why don’t you just combine the 10 .pdfs into a single .pdf then ask away?
@@markoconnor1361 Mark has a good suggestion, but if the 10 combined files are too long then it might exceed the file size or page limitation.
Thank you very much for innovative information'.
Dear Dr. Irshad,
Thank you for the find words. It is a very exciting and confusing time for the academy. Later on, I will also review another chat document interface called chatdoc: chatdoc.com
Best,
Mike
Thanks for your excellent video
Dear Evelyn,
Many thanks for the comment. I appreciate the kind words. I hope the comparison helps. Later on, I will also review another chat document interface called chatdoc: chatdoc.com
Best,
Mike
can you input epub documents aswell or is it only pdf?
Dear Hamza,
I have not tried it with epubs. But if you have epubs, then you can easily convert them to pdf with free online converters. I have used this one before: www.freeconvert.com/epub-to-pdf
Best,
Mike
Which limitations do you see for these types of interfaces regarding whether or not they contribute to humanities research work? Thanks for the video! :)
Dear Automata Learning Lab,
Thanks for the question!
re: limitations
The key limitation is the chat format and the interface's chat limitation with ONE document. You can only interact with the document through questions or chat as the interface tries to makes it easier for the user to control the parameters of the discussion or the chat by limiting it to ONE document. But that is also an intellectual or conversational limit because it does not allow the user to compare an contrast the document with another. A real-life interlocutor might say something like "Oh, that makes me think of this other thing I was reading, and it says..." So, that is both the power and limit of interfaces that allow a user to keep ChatGPT in the "lane" of one specific document.
re: contribution
The ability to chat or talk to a specific document. is the power of ChatGPT because the format of asking "someone" to explain something is very natural and we are already primed to learn in this fashion. I think if you load up a PDF of Plato and Socrates --because of the Socaratic format of these dialogues-- the LLM transformer can almost make it feel like you are chatting with these long dead authors. In that sense, I can see how this can lead to the level of the uncanny in the future. Someone just needs to create some holographic technology and some speech to text interface. Although it could also get very repetitive because their words would be limited to a set vocabulary and pre-loaded content. There can never be a "new" set of thoughts or examples or new patterns of speech. But endless blending of existing speech and semantic parameters. But that is just idle specuation.
re: contribute to humanities research
Yes, sure. Programs like this already allow someone like me to pose new or unusual questions just to see what happens. And as we know, accidents are the mother of invention. These interfaces also creates joy or fun. So, making it fun to interact with an otherwise dry text, for example, might actually make it "more" compelling for a reader and lead to new ways of thinking about the text.
@@UniversityAIed Love the last point about the role of fun in bringing dry text alive! Thanks a lot! Cheers! :)
are there are any other sites you would recommend that can do amazing things with AI
Dear Hamza,
Yes, there are many interesting sites that can help someone with academic research and writing. Have you tried ChatDOC or Paperpal? Paperpal is like Grammerly.
Best,
Mike
Great video, thanks Mike. I don’t really understand how ChatPDF relates to ChatGPT. But if I upload a proprietary pdf to ChatPDF, will that pdf end up being part of ChatGPT’s data and essentially in the public domain? Much appreciated. Mark.
Dear Mark,
ChatPDF is paying for use of the ChatGPT product. So, the authors of ChatPDF have written a program that allows for the user to chat with just one document. It is ChatGPT under the hood, but the interface is ChatPDF.
re: will that pdf end up being part of ChatGPT’s data and essentially in the public domain?
No, I do not think it will be part of public domain. But OpenAI is looking at everything you enter into its system, and the Large Language Model of ChatGPT is "looking" at your uploaded text; however, it is not storing that data and incorporating it into the training data it was exposed to during its training and fine tuning phases. Will it remember that data? I do not think so because it is not part of the training data of the original version (e.g. 3.5 or 4). I think it only remembers it during your engagement with it. And even then it often will lose the thread and go off on tangents that clearly demonstrate it has forgotten what you showed it earlier.
But I could be wrong.
Best,
Mike
Appreciated your efforts, thanks alot
Thank you Omar.
Later on, I will also review another chat document interface called chatdoc: chatdoc.com
Best,
Mike
@@UniversityAIed Glad to hear that Prof.
Thank you for letting me know.
Regards,
Omar
these tools work just with English docs?
Dear Kendall,
Yes, it works with other languages. I just uploaded a French document and it was able to answer my questions in English. So, just like ChatGPT, it is good with other languages. What language did you want to use?
Best,
Mike
Thanks😊
Dear Mufit,
You are welcome. I hope it helps.
Best,
Mike
Hi, love your content. Can you make a video for Sharly AI too?
Dear Simone,
Thank you for the kind words. And thanks for the suggestion. I had not heard of it before. So, that is super useful. I will try it out and post my thoughts.
Best,
Mike
Dear Simone,
I have reviewed Sharly AI. Here it is: ua-cam.com/video/vFXS0z6ap2Q/v-deo.html
Best,
Mike
what about copilot?
Dear Uijwal Chetan,
Do you mean the co-pilot on Bing Chat? Or do you mean Github co-pilot? I don't know much about Github co-pilot but I think that is dedicated to writing code. There is an assistant in Bing that is kind of like a co-pilot that can help you with questions and prompts, but I do not think it allows for uploading a specific PDF. I do like Bing Chat as it runs on a version of Chat GPT 4, so the answers are quite good.
Best,
Mike
keep up the good work
Many thanks A! I hope it was helpful.
M
Queria que humata fosse gratuito ou mais barato para brasileiros, 14 dolares??? Isso dá 70 reais por mês 😢 Algum hacker online pra quebrar isso? 😂
Dear Mila,
I think HUMATA gives you a certain amount of usage for free every month. If you exceed it, then you can always use ChatDoc or ChatPDF.
Best,
Mike
thanks a lot,
You are welcome. I hope it helps.
profs that are afraid of ai are outing themselves lol
Dear Cyberphunkisms,
I am a prof that is afraid of AI! I'm afraid because I know how powerful a tutor and research tool it can be. I am afraid that it might be far easier for a student to ask the AI to do all their work for them instead of struggling to do it themselves. That is the real risk out there now.
Best,
Mike