thanks for the video, great useful resource. i m trying to do the make connection, using a token instead of the oauth connection, when i try to add the "trigger configuration"field, i don't have the "created" option appearing, no option at all actually appear. would you have any idea why? do i have to use the oauth connection airtble < > make? thanks
I think Airtable deprecated some API tokens, so you'll probably have to do oAuth (which is better in my opinion, as you can see every app that has access through Airtable). Hope this solves it (and weird that Make haven't removed the token option yet...)
Excellent detailed OPS. I am experimenting as well 😌 to develop a shorter string. My errors result in double subject text. I will apply some Op changes, as seen in your string to hopefully eliminate my errors. Awesome mentor 😎!
Wow! You must have read my mind - this is exacty what I've been doing in ChatGPT-4 but you've automated it here which seems way better than Bubble. Might you please share the Airtable sheet? Anyway. I'm looking forward to your next video on this topic and hoping you'll share your tools and more how-to's soon. Cheers!
Thanks for this - I am nearly there, but it keeps failing on the Update Chapter Record with this error - The operation failed with an error. [422] Value "" is not a valid record ID.
As I understand it, this means you are not passing the Record ID to this module, so it doesn't know which record to update in Airtable. You should get the Record ID from the Save Chapter to Ebook Chapters module, where you create a new record every time. Once you pass this Record ID, it should work. Hope this helps :)
@@MintyAnalyst Managed to get it working. I ended up deleting the Airtable Module and recreating it, which caused a couple of other issues, but was easy to fix those. I ran a couple of tests and it has a really long way to go. I am wondering if it is running too fast or the prompts need to be improved. On the first test, half of the sentence was missing and in the second test I asked for 3 chapters, The 3rd Chapter started out Chapter 4 - lol. But a great start for sure. I will play with it some more to see what I can come up with. Appreciate all your time and help.
@iainhmunro Yep, it needs fine-tuning of the prompts for sure. As for the responses cutting sentences, I've limited all the OpenAI modules to 300 max tokens to keep it faster/cheaper for testing. If you increase the Token Limits, it should be able to generate everything without cutting sentences.
just start from the step where the outline is being fed to the AI. You can save your outline in Google Docs or Sheets and load it from there, the same way we are adding a topic in the beginning here.
Hello, thank you very much for the great tutorial. I have adopted your scenario 1 to 1 to adapt it to my needs later. Unfortunately it does not work 100% although I have modelled everything step by step as I said. The created EBooks remain open, so it ends in the middle of the sentence.... I hope you understand what I mean... Can you understand where the error may be? Many thanks in advance for your help. Thank you! Greetings Josh
Hi Josh, the only reason for the ebook to end mid-sentence that I can think of is one of two: 1. You are not prompting OpenAI correctly, meaning you are limiting the number of tokens so it can't finish its thought, or your generation prompt is too vague, or you may be using a really old GPT version (not sure if that last one is possible...) 2. You are using the Substring() function in Make on the wrong place and are actually trimming your response before saving it. Those are the only two things I can think of, hope they are helpful :)
Great Video. I am working on same problem of Generating Proposals. I have different prompt in airtable, based on that prompt the ChatGPT generates the Output. One thing is that i want to fallback that if some section is not well written we will write it again. Can we do that?
I've faced similar issues, so I know what you mean... AI is not always good in the generations. For my scenarios, I either review generations manually, or I have another ChatGPT prompt that asks the AI to review and let me know if it looks good. I give it a few examples of good generations and ask it to respond Yes/No if the generation looks okay. You can then save this Yes/No and use it to mark the ones that need further work. I hope this helps 😊
@@MintyAnalyst great can you help me with that? Basically i have column in airtable in which i write please modify the response of prompt 5. But i am unable to update the results. Can i share my example and blueprint, i am stuck in this problem. if you have any linkedin Id?
If there's a more generic question you have, you can ask it here, where others may benefit from the discussion. Other than that, unfortunately, I won't be able to review your blueprint and provide detailed help.
@@MintyAnalyst let me explain the whole . I have airtable in which i am generating whole sales propsal. I have 10 columns for prompt like prompt 1 prompt 2 ,,,, prompt 10 . In which i am writting instruction for chatGPT to generate specific section like intruction background etc. I have succesfully create whole sales proposal. But one issue i am facing that if user review manually and in Column ' Need revision ' he write generate response of prompt 3 again. Then how can i generate that response and update it with already created sales proposal.
Hey, thanks for the comment. I am not aware of any free platform that can do this with the required quality, that's why I am using the OpenAI integration in Make. If you find another one that integrates with Make, let me know, I'd like to try it out 😊
@@MintyAnalyst And in the end we get an article written by humans? Or will the article be detected as written by artificial intelligence? I hope you've checked it out
@IbrahemAb1 If you take it as is and use my generic prompts, it will be easily detected. However, if you research better prompts for your specific use case, and spend some time editing it at the end, results are pretty great. It also depends on which model you use from OpenAI (gpt4 will be better but more expensive than chat-gpt3.5). After all, it's the same performance you get from ChatGPT, just allowing for longer text with a structure.
@@MintyAnalyst Thank you, I wish there was a free site through which we could produce an article with human writing, because all the tools that reformulate the article into a human form change the words in a suspicious manner and put in strange and inconsistent words and meanings.
If you copy-paste my scenario, leave the short generic prompts, and publish the first generation with zero editing, then it's exactly that. BUT, if you spend the time to craft the right prompts, and then spend the time to review and edit the generated text, I don't think it's spam, it's just a great time-saving tool.
@@MintyAnalyst The reality is that AI-generated content simply isn't very useful in at least most B2B marketing programs. You need to be much sharper, better targeted, and informative that AI is capable of. It'll frankly never get there. Not sure about B2C, it may be different.
Hah 🤣 Yep, AI is not THAT good yet, but I've found a lot of use cases for this, especially if you spend some time to fine-tune the prompts. In the end, it's always meant to be a first rough draft, not a finished product. But yeah, if you want to just copy-paste the scenario, do zero work, and get amazing results, it will be a wasted $1 😅
@@MintyAnalystIt will be a wasted $1 anyway in most cases, as you would lose even more, your time, as AI wouldn't be able to produce a proper result due to inability to track the context over large amount of text, and produce poor quality text due to constant losing of focus if the context is explicitly specified (which, in turn, can't be NOT large).
@@trevoro.9731 I don't fully agree. Yes, AI is repetitive when no context is provided. But you can play around with that (not even mentioning that context windows are hugely increasing all the time). You can pass the summaries of all chapters and ask it not to repeat itself in the current chapter, you can pass the list of chapter titles, and ask it not to cover topics in other chapters, etc. There are countless ways to "help" AI be better at generating content. It all comes down to understanding that there's no one-click magic solution. You need to put in the work to get good results. As the saying goes, 'Sh*t in - sh*t out.' However, I still believe with the proper understanding of GenAI and time spent fine-tuning prompts and workflows, the benefits are huge. For example, AI helped me come up with the scenario in this video. I think people who straight out dismiss it are missing out a LOT 😉
@@MintyAnalyst I tested my own ways of forcing LLM to generate content, it resulted in constant loss of context, prioritizing trained data which make no sense for the content, and (after fine tuning) sometimes delusions of neural network. It is unlikely that the service does even that (as it would require a lot of time).
thanks for the video, great useful resource. i m trying to do the make connection, using a token instead of the oauth connection, when i try to add the "trigger configuration"field, i don't have the "created" option appearing, no option at all actually appear. would you have any idea why? do i have to use the oauth connection airtble < > make? thanks
I think Airtable deprecated some API tokens, so you'll probably have to do oAuth (which is better in my opinion, as you can see every app that has access through Airtable). Hope this solves it (and weird that Make haven't removed the token option yet...)
"...with One Click" my a$s 😄😠😑🤡
Well, it takes a lot of pre-work and prompt fine-tuning, but once you have those, it's literally one click 🧑💻
Excellent detailed OPS. I am experimenting as well 😌 to develop a shorter string. My errors result in double subject text. I will apply some Op changes, as seen in your string to hopefully eliminate my errors. Awesome mentor 😎!
Hey, thanks, I am happy the video was helpful! Good luck with your experiments! 😊
Wow! You must have read my mind - this is exacty what I've been doing in ChatGPT-4 but you've automated it here which seems way better than Bubble.
Might you please share the Airtable sheet?
Anyway. I'm looking forward to your next video on this topic and hoping you'll share your tools and more how-to's soon.
Cheers!
Hey, thanks for the message, and positive feedback. Here's the Airtable base: airtable.com/appe4ZJX8t2utfBNW/shroUJ9MUjcdnnQZG
I wish I saw this as I built from scratch
@@iainhmunro sorry, just figured out how to export make scenarios to json today 😅
Thanks for this - I am nearly there, but it keeps failing on the Update Chapter Record with this error - The operation failed with an error. [422] Value "" is not a valid record ID.
As I understand it, this means you are not passing the Record ID to this module, so it doesn't know which record to update in Airtable. You should get the Record ID from the Save Chapter to Ebook Chapters module, where you create a new record every time. Once you pass this Record ID, it should work. Hope this helps :)
@@MintyAnalyst Managed to get it working. I ended up deleting the Airtable Module and recreating it, which caused a couple of other issues, but was easy to fix those. I ran a couple of tests and it has a really long way to go. I am wondering if it is running too fast or the prompts need to be improved. On the first test, half of the sentence was missing and in the second test I asked for 3 chapters, The 3rd Chapter started out Chapter 4 - lol. But a great start for sure. I will play with it some more to see what I can come up with. Appreciate all your time and help.
@iainhmunro Yep, it needs fine-tuning of the prompts for sure. As for the responses cutting sentences, I've limited all the OpenAI modules to 300 max tokens to keep it faster/cheaper for testing. If you increase the Token Limits, it should be able to generate everything without cutting sentences.
@@MintyAnalyst Many thanks. I didn't even think about the token limit, so that makes sense now.
How can you create (Created Date and Time) and Last modified
You have such fields as pre-defined options in Airtable. Add them there, and they will auto-populate and auto-update.
@@MintyAnalyst thanks. I have paste my problem i am solving can you please help me with that?
@@hammadyounas2688 not sure what you mean, where have you posted it?
@@MintyAnalyst you have replied there. My problem of proposal writing.
Thanks for this video. That's amazing. Direct from Brazil!
Hey Fabiano, glad you enjoyed it! Another Make.com video is dropping next Thursday 😊
I love this, but what if I already have an outline. How would I start this process from the outline?
just start from the step where the outline is being fed to the AI. You can save your outline in Google Docs or Sheets and load it from there, the same way we are adding a topic in the beginning here.
@@MintyAnalyst Sweet, that makes perfect sense now that you have explained it.
Hello, thank you very much for the great tutorial.
I have adopted your scenario 1 to 1 to adapt it to my needs later.
Unfortunately it does not work 100% although I have modelled everything step by step as I said.
The created EBooks remain open, so it ends in the middle of the sentence.... I hope you understand what I mean...
Can you understand where the error may be?
Many thanks in advance for your help. Thank you!
Greetings Josh
Hi Josh, the only reason for the ebook to end mid-sentence that I can think of is one of two:
1. You are not prompting OpenAI correctly, meaning you are limiting the number of tokens so it can't finish its thought, or your generation prompt is too vague, or you may be using a really old GPT version (not sure if that last one is possible...)
2. You are using the Substring() function in Make on the wrong place and are actually trimming your response before saving it.
Those are the only two things I can think of, hope they are helpful :)
@@MintyAnalyst hello, thanks for the feedback, i have now re-installed everything and everything is working now. thanks
@@Josh-RO perfect, happy to hear that!
Great Video. I am working on same problem of Generating Proposals. I have different prompt in airtable, based on that prompt the ChatGPT generates the Output. One thing is that i want to fallback that if some section is not well written we will write it again. Can we do that?
I've faced similar issues, so I know what you mean... AI is not always good in the generations. For my scenarios, I either review generations manually, or I have another ChatGPT prompt that asks the AI to review and let me know if it looks good. I give it a few examples of good generations and ask it to respond Yes/No if the generation looks okay. You can then save this Yes/No and use it to mark the ones that need further work. I hope this helps 😊
@@MintyAnalyst great can you help me with that?
Basically i have column in airtable in which i write please modify the response of prompt 5.
But i am unable to update the results.
Can i share my example and blueprint, i am stuck in this problem. if you have any linkedin Id?
If there's a more generic question you have, you can ask it here, where others may benefit from the discussion. Other than that, unfortunately, I won't be able to review your blueprint and provide detailed help.
@@MintyAnalyst ohh but here i am unable to attached image and other things.
@@MintyAnalyst let me explain the whole .
I have airtable in which i am generating whole sales propsal.
I have 10 columns for prompt like prompt 1 prompt 2 ,,,, prompt 10 .
In which i am writting instruction for chatGPT to generate specific section like intruction background etc. I have succesfully create whole sales proposal. But one issue i am facing that if user review manually and in Column ' Need revision ' he write generate response of prompt 3 again. Then how can i generate that response and update it with already created sales proposal.
very good video . where can i find this template please
Hi there, unfortunately, only Team and Enterprise plans can share make.com scenarios as templates, so there's no way for me to share it 😐
You could export the blueprint..
@jeanluclaval thanks. not sure how to do that, but will take a look 😅
great tutorial! took me a little while to do it but awesome. thanks a mil man.
glad it was helpful! 😊
Thank you very much for this valuable tutorial! you can share the airtable and Json file with us please
Hi there, here's an Airtable link: airtable.com/appe4ZJX8t2utfBNW/shroUJ9MUjcdnnQZG/
However, I am not sure what JSON file you are referring to 😅
@@MintyAnalyst Thankyou 🙏
@@MintyAnalystPlease make an effort, I really need it
I have no idea what you are referring to... 😅 what do you need?
Finally figured the JSON stuff out 😅, you can get it here: mintyanalyst.com/free-ebook-automation-system-in-make-com/
You are a good man! 3:30 am and I'm happy as a hobbit with a habit! Thanks!
Thanks for the comment, I'm glad you like the video 😊
GPT Chat does not work. It requires a paid plan in order to use Open AI. Can it not be replaced with a free platform?
Hey, thanks for the comment. I am not aware of any free platform that can do this with the required quality, that's why I am using the OpenAI integration in Make. If you find another one that integrates with Make, let me know, I'd like to try it out 😊
@@MintyAnalyst
And in the end we get an article written by humans? Or will the article be detected as written by artificial intelligence?
I hope you've checked it out
@@MintyAnalyst
Will the text after here look like it was written by humans when examined with tools that detect writing with artificial intelligence?
@IbrahemAb1 If you take it as is and use my generic prompts, it will be easily detected. However, if you research better prompts for your specific use case, and spend some time editing it at the end, results are pretty great. It also depends on which model you use from OpenAI (gpt4 will be better but more expensive than chat-gpt3.5). After all, it's the same performance you get from ChatGPT, just allowing for longer text with a structure.
@@MintyAnalyst
Thank you,
I wish there was a free site through which we could produce an article with human writing, because all the tools that reformulate the article into a human form change the words in a suspicious manner and put in strange and inconsistent words and meanings.
spam, you're writing spam.
If you copy-paste my scenario, leave the short generic prompts, and publish the first generation with zero editing, then it's exactly that. BUT, if you spend the time to craft the right prompts, and then spend the time to review and edit the generated text, I don't think it's spam, it's just a great time-saving tool.
@@MintyAnalyst Very much disagree, but that hardly matters.
@Charles-yq8vv I get what you mean, but at the end of the day, AI is here to stay, so we either adapt or we get left behind
@@MintyAnalyst The reality is that AI-generated content simply isn't very useful in at least most B2B marketing programs. You need to be much sharper, better targeted, and informative that AI is capable of. It'll frankly never get there. Not sure about B2C, it may be different.
How to pay 1$ for a book-sized collection of BS ...
Hah 🤣 Yep, AI is not THAT good yet, but I've found a lot of use cases for this, especially if you spend some time to fine-tune the prompts. In the end, it's always meant to be a first rough draft, not a finished product. But yeah, if you want to just copy-paste the scenario, do zero work, and get amazing results, it will be a wasted $1 😅
@@MintyAnalystIt will be a wasted $1 anyway in most cases, as you would lose even more, your time, as AI wouldn't be able to produce a proper result due to inability to track the context over large amount of text, and produce poor quality text due to constant losing of focus if the context is explicitly specified (which, in turn, can't be NOT large).
@@trevoro.9731 I don't fully agree. Yes, AI is repetitive when no context is provided. But you can play around with that (not even mentioning that context windows are hugely increasing all the time). You can pass the summaries of all chapters and ask it not to repeat itself in the current chapter, you can pass the list of chapter titles, and ask it not to cover topics in other chapters, etc. There are countless ways to "help" AI be better at generating content. It all comes down to understanding that there's no one-click magic solution. You need to put in the work to get good results. As the saying goes, 'Sh*t in - sh*t out.' However, I still believe with the proper understanding of GenAI and time spent fine-tuning prompts and workflows, the benefits are huge. For example, AI helped me come up with the scenario in this video. I think people who straight out dismiss it are missing out a LOT 😉
@@MintyAnalyst I tested my own ways of forcing LLM to generate content, it resulted in constant loss of context, prioritizing trained data which make no sense for the content, and (after fine tuning) sometimes delusions of neural network.
It is unlikely that the service does even that (as it would require a lot of time).
@@MintyAnalyst If they just dismiss it outright here, they would save at least 1 $ without missing anything.