today's video is going to be a bit of a mixed bag we're going to continue trying out open ai1 uh focusing mostly on the mini version uh today because we are playing here over on cursor I've been testing out different cursor rules to see if that has any effect on my productivity my processes and I've been kind of trying to up my prompting game by using more XML tags more structure and to do that I've been trying to leverage cursor here to actually help me rewrite my prompts into this yeah you can see here XML tags to hopefully get a more yeah precise instruction so yeah I think we're just going to dive in we have some examples we're going to do so yeah let's just get started so over the weekend I've been also trying to look out for what other people think of using let's say 01 mini for coding uh maybe on cursor so I found this Reddit posted that I thought was pretty good and kind of fits my view of 01 mini to after trying it out over this weekend so I thought we can just go through a few of the points this uh who's afraid of 138 has made here I thought it had some good points here so you can see the fast few days I've been testing 01 mini uh in cursor compared to 3.5 yeah which has been a Workhorse model that has been insanely consistent and useful that is kind of my IDE too so the verdict is claw 3.5 sonets is still a better day-to-day model and that is kind of my takeaway too just because of speed and reliability and yeah that is my idea right but I'm open to change right and I've been trying out 01 mini over this weekend so I just wanted to go through some of the pros he found and some of the cons and see if I agree here so the 64k output tokens uh that is big right so if you compare I think a CLA 3.5 maybe has 8K output tokens so that's eight times more output you can print right in cursor so that is nice when you do these composer features and you won't like create bunch of files that we're going to look at later uh so you can see uh if your prompt is good it generally can do a large refactor re architecture job in two three shots yeah I think that's good uh I'm not going to read example so in general this was quite a large of reflecting job and it do well large output context is a big part of facilitating this I think that's pretty cool uh so here he has listed some cons that I kind of agree with so you have to be very specific with with your prompt and that is kind of what I dived into these XML tags this weekend to try to be even more uh instructive uh than I've been with clo 3.5 son to see if that can improve something on when using 01 mini in cursor right and you can see due to Long thinking time you pretty much need a perfect prompt because it's annoying right if you don't have good enough instructions and you waste time by waiting for these 01 models to do their Chain of Thought and and H there was something wrong right uh so he is working with 01 you have to do everything one shot uh I have not been doing that but let's see uh limited chats yeah that's a huge limit there of course but this is early 64 outcut that has to be that is not a con right uh I wish son 3.5 has uh much more output tokens yeah if Sonet had 64k that would be pretty cool uh verbos yeah cursor implementation is buggy this is important so sometimes there's no text output only code I've seen this too and I wonder this if this is something to do with um cursor backend uh prompting I'm not quite sure uh so I've been trying to do kind of my own let me show you here I did this uh let me see here yeah so I try this 01 rule so ignore all Chain of Thought step by step prompting when using cursor ignore chain of thought step by step prompting me using cursor chat but I don't know we might try it but I don't think it's going to work to be honest so we can see he has 01 mini CLA five son conclusions so if you're doing a massive refactoring job or gree and filling a massive project use 01 mini combination of deeper thinking and massive output token limits means you can do things one shot I haven't tried that too much so I can't really reflect on that so if you have a collection of smaller task CL Sonet is still the king so be very specific and overl Bose in your prompt to1 mini describe as much as your task in detail as possible it will save you time because this this is not a model to have conversations or fix small bugs it's a Ferrari to the Honda at is Sonet so yeah I've been switching models right so I maybe I do like one shot with 01 mini and then I kind of switch to 3.5 Sonet so that's the way I've been solving this but I just thought this post was pretty interesting and it kind of confirms my uh comparisons with 3.5 Sonet 2 so I just wanted to go through this because it's ni to get like a confirmation of what you've been feeling too right so I think we're just going to dive over to cursor now test some rules do some prompting with XML text so let's just head over there so if you see here on cursor in the settings you see we have something called rules for AI right and here you can kind of set these general rules that will apply for all your sessions right but we can also include a DOT cursor rules files if you didn't know that and in this uh project we kind of have these uh different rules so always assist the user and suggest XML rapper tags to improve the prompt uh the user rights right and we're going to the suggestions must help must help the user make it as clear as possible for an llm to understand the scope of the prompt so these are the rules I have been testing out and I just wanted to show you kind of how I've been playing around with this and how I've been using cursor to kind of write my prompts instead of doing it like yeah just in the chat box right so if you're planning to try this for yourself uh you got to kind of look at down in the right corner here there's something called uh cursor tab so here we have a hacker news. mark MD that's kind of a markdown file so in this cursor tab here you can see sometimes it says uh disable for markdown uh so if this is kind of checked off you can see we have this cursor tab that is kind of not available here so uncheck this so you can see we can use this tab so you can see when I start writing now right we get some Auto suggestions right if you wanted to use that if not just go back and just start typing so let me just come up with a prompt here and we can try to show you kind of the way I've been playing around with this okay so here you can see my prompt now I want to create a terminal app that fetch the top 10 posts on Hacker News and display them in a retro terminal style uh I want to scrape The Hacker News website for this to post using bs4 I want to use the rich library to style it the terminal app must include the links to the post and the title of the post the code should be in Python make the code modular easy to understand and of course be effective and secure as possible this I kind of wrap in Project the action for llm is to execute the project above okay so that is a prompt I used to use right so what I've been testing out here I just uh select all contrl K so here I just go wrap the prompt in XML tags to make the instructions as clear as possible for the llm I just do submit uh edit right and we are on CL 3.5 Sonet here that's fine and now this is going to rewrite this it's going to add all the XML tag it thinks is the best possible instructions here to kind of get the most out of that so you can see here right so it right so let's just accept this and take a look now so you can see we have description we have description requirements and we have some requirements here kind of wrapped in this XML tags we have action step by step here right okay and yeah I think this is fine so let's go ahead and try this out now so let me just open our new cursor here and let's test this prompt using open ai1 mini I see I forgot to add something I've been also focusing on lately let me uh select all this and let's do include our instruction to the llm to include the folder structure of the project with old folders and files so this is also something I've been playing around with lately because uh using o1 we have this big uh 64k output token so it's super easy to kind of recreate folder structure using the composer feature so you can see here now uh we added a folder structure I don't think I want it like this so let me rephrase this prompt here so let's do include a step to print a pH structor with XML tags just the instruction so I just wanted to add this into the steps here right for Action let's see if we can do that okay good now we have it here so just accept that let me save that and now let's try to run this so I just open a new directory here let's paste in our prompt right uh now I want to select 01 mini uh because these are very precise instructions right so let's just try to run uh in the chat here first if we get the Fage structure we're going to take that and we're going to go to the composer and try to create all the files before we add the code right okay that was pretty quick so now you can see we have our project structure so I'm just going to copy this control I right paste in this and I've kind of played around with uh so this is kind of my instructions for generating the folder structure so so you're an L liit access to generating uh folds and filers in the current working directory generate all files and folders so yeah we have some action here so I'm just going to call copy this to head back to our composer and just paste in this so let's try this now so we are on the 01 mini right okay so you can see we have some placeholder here we have the requirements so let's just accept all this and if you go back here now you can see we have all our files right and then we can just start adding here so we can do requirements let's just add this right we have our scraper dop good we can apply the code here accept save we have uh the display so now we have all files ready right should be pretty easy we have main let add this install the dependencies okay install requirements good and then we can run it okay uh that was a bit strange you can see it's running but we missing the posts here uh so let's try to do like a quick followup here and see if we can fix this so I'm just going to go the app is running but we don't see any hacking News Post in our terminal but now I'm going to select CLA 3.5 control enter to kind of use the code base so this is what I've been doing I've been switching between 01 mini o1 preview and Claw 3.5 depending on the use case right so we want to go to our main we want to add some new code here right okay uh scraper let's add some new code here uh let's see now so let's clear this let's run it and boom we got it so here is our app right top 10 Hacker News Post uh we have this link we can follow right okay boom we got it so that was pretty cool so yeah seems to be working pretty nice uh our F structure was good our initial prompt from 01 mini was pretty good good and then we switch to claw 3.5 for some debugging and boom we got it so yeah so yeah that was pretty good I'm happy with this I just wanted to do one more example to show you kind of how I've been using 01 mini to create these folder structures because that's been pretty interesting right so here you can see we have a huge folder structure right and you can see here I kind of took my prompt in uh in the end here so you can see the action is create a specific five Direct and files in the current work directory for each file include a basic placeholder content that represents it purpose or structure so if we grab this full prompt here so this is a pretty big folder structure right so if we grab all this uh let's head over to yeah we can do just the same folder here so let's head control shift I into the composer let's remove this old one and now you can see we paste in like the full structure here and here is where the kind of the big output window comes uh 64k output token window comes into play so you can see in the bottom left here I guess you can't see because of my face but we are selecting o1 mini and when we run this now hopefully we can create this big folder structure uh with files and folders so let's see now so you can see now it's writing all the files right so this is pretty big it's that's a lot of files to create right and it was pretty fast when it first got going here so if we accept all this now right accept all we go back here boom we got it so we have our back end right uh with all the files middleware chat controller Roots Services open I service our front end our app page layout Styles Global CSS components right uh so what we had to do now is just fill in this because we already have created all the files we could just since you can see they have this placeholder code so if we had the code ready now we can just apply this to all the files so that means that we didn't manually have to create all the file Series this is a big Advantage I have found with using 01 mini uh using this composer just because of the big uh output token window right or output tokens so this is a bunch of files I'm sure we can do even more if we try to so I think this is a very good use case so now if we want to chat we can just switch to CL 3.5 and start adding code here right I just think this is very interesting and this is definitely something I will be implementing into my workflow so the final thing I thought we can do today is actually do something that is impacting me because this is a feature I've been wanting for my own website so let me just show you now so this is the setup I have for my website if we head over here now I have a section called videos so here I kind of upload all my latest videos right uh but the way it's set up now if we we go here I have to go into the code here right I have to go to website I have to find my Json structure down here somewhere where is it yeah you can see here uh I have to fill in this small Jon here that has the title description and iframe that is kind of the embedded video you can see here and it's very annoying to do this manually so I kind of want to create a script that can just maybe just link the URL and the title and justun the script in the terminal and this will generate this kind of structure uh and I kind of want open a i API to write the description based on the the title right so let's see if we can actually use cursor open a i uh1 mini I think and maybe some 3.5 to actually create this uh pipeline for me and this is kind of my raw prompt before I add the XML tag so I want to create a pipeline to update my react website it's easier to use than the current method now I have to go into the code in the website to find a code where latest videos are update the section uh I want to add a nice pipeline I can use to update the latest video section on my website I want to be able to add a YouTube video by copying the URL pasting into a script that adds the video Title by the video into a script that adds the video to the website the script should be able to run with the following command so I just want to add video URL and title to write the description for the YouTube video I want to use llm to generate a short description based on the title uh used open a i API the generated Json structure must be fed into website JS to update the website I pasted in some documentation so that's the gp40 mini implemented features above into my re website project I will add website JS as context so now I'm just going to select all contrl K and let's wrap this into XML tags so let's just wrap my prompt in XML tags to make the task as clear as possible so let's just try this and see what structure we get now okay so let's accept this and take a look here so now we kind of have an objective we have a current method okay we have the script we want the process and the description generation okay that's pretty good and we have the implementation request so an action and some context I think we just going to try this and see if this actually works so let me just copy this head over yeah to our uh directory here that we have all the files so what I want to do now is open up our composer right uh I want to add the website as context I want to add maybe like the folder so that's going to be Source I think and let's paste in our prompt here now so I think we just going to try this again I want to select 01 mini here for the first iteration so let's run this and see where this takes us okay okay so now you can see we have our EnV file we have our videos Json so this is kind of our where we're going to store our examples if we look at the new video section now we're going to load from our Json file I think and the add video script uh if we open up this I think we're going to run this command uh yeah we're going to include our video and our title okay that's interesting we need to install some dependencies yeah I think this looks pretty interesting and it looks pretty good so what I'm going to do now is I'm going to accept all this I'm going to implement all the code I'm going to set it up and then I'm just going to try it and see how this actually work uh yeah let me do that and I'll come back to you when we are ready okay so now I kind of want to show you how this works so if we head to my website now uh you can see we have this video here in latest videos right so I just did a quick test looks superb right so if we go here now you can see uh we are using open AI gp40 mini to generate a concise and engaging description from based on its title right so we have some examples here on kind of the ey framing right and yeah I'm not going to go too much into the code but this is going to be you can see new videos will be inserted here by the ad. video.js script perfect so how this works now is we can do so what we can do now node uh at video at video.js right we can do D- URL and then I can head over to let's say my YouTube channel uh let's just copy this URL here I can go back to cursor paste in this URL I can do d-h title uh let's do a string and I can go back here let's just what was it copy this title and paste it in here okay that kind of missed so let me go back and fix that and we need a string here uh okay so I'm going to add this so generating description using open AI boom okay so you can see this got slotted in here it got saved if we go to our website now boom we got a second one so that was super nice pipeline for me let me just go grab another video here so AI app of the week 2 I can just do node ad videojs URL uh okay so I have the title boom generate description add it here go back to my we site boom we have another video that saves me a lot of time so this is a superb use case so I'm super happy with this this is going to save me a lot of time so yeah combining uh 01 mini with kind of the The Prompt we created with all the XML tags and we use some CL 3.5 for some simple debugging how much time did I spend on this that's maximum 10 minutes so that's a superb integration into my website by adding like a a new feature that saves me a lot of time when updating new videos so this is something I'm going to just Implement and deploy right away so that's pretty cool right uh yeah super happy with this and the way it turned out and combining all of these things I've been playing around with this weekend I think this yeah kind of helped me with my productivity so I hope you found some of these things interesting uh if you want to access all of these prompts I have here uh just follow the link in the description I have kind of open public repo where I will be putting out all of these yeah the small prompts here I've been using so just go check it out if you want to copy maybe my folder structure instructions or something like that also if you want to shout out something in my YouTube video some GitHub project you have you can just go to my website just go to Services I have this YouTube video shout out I'm not sure about the price yet so you can kind of buy if you want to do some shout out maybe on your GitHub repo other things you want to shout out some small startup you're working on just send me a request here and maybe we can do something so or I I can put out the cursor rules too if you want to copy this but uh just go play around try out different things with 01 if you have access if you have cursor you should have access to at least a few uh requests right uh but I think still there's a lot of things we need to explore here when it comes to using the new models uh but uh daily driving is going to be CLA 3.5 maybe some GPT 40 but I'm going to keep experimenting using uh using the 01 pre uh not not so much the o1 preview mostly I'm going to be using 01 mini I think but I'm going to try something with the preview too and it was pretty cool to see uh what other people have been using on Min I've been watching some videos this weekend there's a lot of cool stuff out there uh but yeah I kind of agree with this CLA 3.5 son it is still uh the better day to-day model for me for now uh but the 64k output tokens in 01 mini makes it super easy when we want to do this uh big folder structure so that is something I'm going to using it for so I'm going to keep experimenting let me know in the comments if you have any cool things you have found out about using o1 yeah I hope you took something away from this video maybe you learned something new maybe something you want to try let me know in the comments give it a like if you found it interesting and yeah have a good day and we speak soon