Vision-based Web Scraping with the New GPT-4o model in Make.com

[@yangpten]

📚 Free AI Course 👉links.dumplingai.com/ai-productivity-course 🤖 Need help scaling your business with AI Automations? Book a free intro call: links.dumplingai.com/ai-services

[@idontexist-satoshi]

For future reference, your JSON error was because you had a `,` after the last field. You don't do this if a new field isn't declared after it. {"field_1": "content", "field_2":"content"}

[@yangpten]

Super helpful!

[@jambogamer-je2nf]

ironic as to how he could just verify his code onto chat gpt, in retrospect. but for the purpose of keeping the video short he rushed it a little. its amazing how the human genome has come with the best possible capacity for pattern recognition, but unfortunately it heavily relies on the need for consistency, which in itself severely lacks. (the best possible until now at least). now any pattern for the omni ai is like a toy, whether it being sequencing the human genome or calculating the trajectory of our planet in relation to the black hole closest to our solar system.

[@darinpatterson9485]

Just FYI, you can use Make's OpenAI Analyze Image module to do this a bit more simply. When choosing the model, instead of using the dropdown click on the "map" button and then type in gpt-4o

[@yangpten]

Good call - I forgot about this, have been playing around with the playground too much!

[@manojBadam]

This is a great example, it opened up few usecases for us.. Thank you for sharing this

[@yangpten]

Glad it was helpful!

[@TheMeanor]

Cool video - thank you for sharing! It worth to be mentioned that image-based scrapring is extremely slow vs accessing selectors directly. I guess it's the way to go if speed and cost isn't an issue.

[@yangpten]

Great tip!

[@docdanoderm]

I really wanted to like this automation, but the $600/year price tag for the screen recording integration is too much.

[@yangpten]

You can always use the HTTP module built into Make.com as a free option if you just need basic scraping :)

[@punktkommastrich007]

Why dont you use rss-Feed for the Data?

[@agi.kitchen]

you can hire me to build you an agent that doesnt rely on third party tools and has default mechanisms (so more reliable), and can pay me at a cost over time

[@pushREC]

Dope workflow! Another sick usecase is scraping amazon product pages which are very visual and hard to scrape 👀

[@yangpten]

Agreed!

[@moosa3956]

hey awesome video, when you were trying to parse the json with sample data, it probably failed because of the "trailing comma". So on the last line, you don't add a comma after the value. ``` { "property1": 123.2, "property2": 123.34, "final_property: true } ``` Hope that helps for the future

[@yangpten]

Thanks!

[@SalyersAI]

Awesome video! Can't wait to play around with my scraping models with this method!

[@deeplypresent]

Hey man great video. The issue with the JSON validation was the trailing comma on the last item. Ran into that myself a few times!

[@yangpten]

Thanks for the info!

[@RolandoLopezNieto]

Great video, tons of ideas, thanks.

[@yangpten]

Thanks for watching!

[@rezashah22]

Thanks for the video. Could you please provide some guideline how to scrape an ecommerce site which has multiple product pages. Rather than once single page scraping!

[@6mystique]

oh my gosh, i grasp olny thats powerful tool but unfortunately I don't understand the usecase. just my mind is a bit closed in this area, but honestly I wish to understand this and make money everyday on the crypto for example. ehhhh. thank you for video !

[@nastastic]

I think this is too expensive to run and scale. Better off making a script with puppeteer or scrapysplash

[@aimattant]

Just for non coders and businesses who do not want to waste time and set it up instantly and have money to waste. There are two sides of rhe AI story. But definitely doing it the pyhton way and adding the GPT api and running it on replit, and then setting up a white lable software for a business is the way to go. I plan to full stack it like that for businesses soon with AI autoamtions in most important business areas.

[@Comic_Book_Creator]

thank you .. great example

[@bennguyen1313]

How does it do with OCR? For example, if I handwrite a 3x3 table of numbers, scan it to a PDF, then ask GPT-4o to sum the numbers in the 2nd column..

[@aamironline]

Other issue with the vision based scrapping is sometimes page's scroll handling may not be favorable, and prevent full page screenshot.

[@yangpten]

Yes that's another downside with vision-based scraping (especially infinite scroll type pages)

[@jaysonp9426]

It feels like all of the no coder channels have no idea what its actually like to build a real solution. GPT4o is not accurate enough to do...very expensive OCR.

[@yangpten]

Agree it’s too expensive for many use cases for now, but it’s an easy and fast option especially if you are already using LLMs in your build. I’d recommend running some evaluations on dedicated OCR systems. Their accuracy often leaves much to be desired 😂

[@justinalexanderbrown]

Beautiful singularity golden boom beautiful ❤️❤️ 100% pure energy love 💕 2024 love you too baby

[@IshaqKhan010]

how much it will cost if thousands of urls to be scrape daily?

[@alexo7431]

wewll explained, thank you for the idea. Too bad that openai's playground costs money

[@yangpten]

No worries!

[@agi.kitchen]

as someone who has workedin automation, what are you gonna do if they change their interface, since everything is being read with json? Why not just read it as an image then with selenium or bs as default in case they change their UI you dont have. to change your code and users dont experience a hiccup in the prices, or even better, if it changes during. a critical time and you have a bot depending on the data, if you can't capture it with the openai action, you stil capture it. I am sayin that cuz i build ai agents for a living and it's failed ever so often

[@yangpten]

We're reading the page as an image in this video. The JSON is the output. Having the standard scraping methods as a default works in cases where the pages you are scraping fall within a fixed subset.

[@agi.kitchen]

@@yangpten totally fair, and i get that, i meant more if hte gpt api fails, which it does intermittently. anyhooo awesome video!!

[@Fullhaura]

How do you get Dumpling AI available in Make please ?

[@yangpten]

You need to request access on dumplingai.com

[@NeoFlashXx]

how much does each fetch cost?

[@d.d.z.]

Same question

[@yangpten]

A few cents

[@smilli6415]

how can i get start up leeds using chat gpt

[@iainhmunro]

Hi Yang - how do i get access to Dumpling Module for Make - I sent in a request a few weeks ago

[@yangpten]

Let me see if I can fast track you

[@iainhmunro]

@@yangpten That would be appreciated - I could help provide additional feedback

[@aldyreal]

This is the most expensive OCR ever.

[@tahooo1402]

hi, nice video, but i can´t find dumpling ai inside make?!

[@yangpten]

You need to install it via www.dumplingai.com/

[@4tofromcafe]

Seems complicated. What are benefits over just using a quck screen grab and pasting it in to chatgpt and asking your question?

[@yangpten]

This is automated - You can scrape hundreds/thousands of pages. Doing that manually will take lots of time and effort.

[@HamedMaddahi]

Scale and scheduling the operations

[@rulofmg]

bruh you dont even show the results. All we see here is bunch of fail and errors. how appalling it is to push your own product that doesn't even work. update your service first then show us in nice format. failed marketing

[@yangpten]

The results are shown multiple times? Did you watch the video?

[@Experternas]

you should use AI to make your voice more human-like