Interesting to see this from the start. I have used massive prompts like this to try to make big plans, and its good for clearing up some thoughts in my head, but it turns into an absolute disaster as soon as I try to make it give me an actionable plan. First it will make mistakes, which I think we are about to see (I'm at 23:30 and you have immediately spotted that its picked a dodgy version for express). And second when you go down a rabbit hole of clarifying and refining any part of the massive plan it just turns into a massive scroll fest trying to move back to the original plan, and have a conversation with it at the same time. Usually the plan will diverge and then you have to somehow think of the new prompt that starts from where you are at, but still has the full detail. But thats not very fun. Instead of getting the AI to do the work for you, it ends up becoming work to try to write a big prompt and manage the whole process along. Thirdly, doing it all in big batches like this, where it has suggested a dozen or so npm packages, or will generate a component that has all the considerations already implemented, is not how I develop. So it means there is a massive burden of trying to fit the whole thing into my brain in one go, and reason through what its generated, and the git commits are just in big blocks of code, rather than iterating adding each bit feature by feature, layer by layer. So in the end I've stopped attempting to actually work with ai like this. I just generate and move forward one function or small feature at a time. So that I can control the direction of the conversation and keep a mental model of what we are doing, and commit at each step. And to be honest, chat gpt seems to have really gotten worse over the last few months. It's at the point now where it will lose context of the conversation within one or two replies.
Thanks for your message. Yer I've definitely fallen into this trap but as you may see from the long number of videos it took a while to really realise this. And the last 10 - 15 or so videos have been me working with AI to re-write code and refactor things. I'm almost at the point now of being able to write some new functionality and have a refined process for doing so that I hope to test out. But yes, working in smaller steps is definitely a requirement, as well as constantly reminding AI of how we want to do things, packages we are already leveraging, etc. Hopefully I find the right mix and can actually speed up rather than get bogged down by AI going in circles like it has been for a lot of my videos.
@paul-towers ah ok it sounds like you are in the same position as me. I am so much more productive with ai but honestly getting drained by the constant massive swing between speaking to a hyper intelligent god and a drunken toddler 😂
![[Bad] - YouTube ВСЁ? (победа VK и прощание с подписчиками)](http://i.ytimg.com/vi/pQ1bCOW5270/mqdefault.jpg)
Interesting to see this from the start. I have used massive prompts like this to try to make big plans, and its good for clearing up some thoughts in my head, but it turns into an absolute disaster as soon as I try to make it give me an actionable plan.
First it will make mistakes, which I think we are about to see (I'm at 23:30 and you have immediately spotted that its picked a dodgy version for express).
And second when you go down a rabbit hole of clarifying and refining any part of the massive plan it just turns into a massive scroll fest trying to move back to the original plan, and have a conversation with it at the same time. Usually the plan will diverge and then you have to somehow think of the new prompt that starts from where you are at, but still has the full detail. But thats not very fun. Instead of getting the AI to do the work for you, it ends up becoming work to try to write a big prompt and manage the whole process along.
Thirdly, doing it all in big batches like this, where it has suggested a dozen or so npm packages, or will generate a component that has all the considerations already implemented, is not how I develop. So it means there is a massive burden of trying to fit the whole thing into my brain in one go, and reason through what its generated, and the git commits are just in big blocks of code, rather than iterating adding each bit feature by feature, layer by layer.
So in the end I've stopped attempting to actually work with ai like this. I just generate and move forward one function or small feature at a time. So that I can control the direction of the conversation and keep a mental model of what we are doing, and commit at each step.
And to be honest, chat gpt seems to have really gotten worse over the last few months. It's at the point now where it will lose context of the conversation within one or two replies.
Thanks for your message. Yer I've definitely fallen into this trap but as you may see from the long number of videos it took a while to really realise this. And the last 10 - 15 or so videos have been me working with AI to re-write code and refactor things.
I'm almost at the point now of being able to write some new functionality and have a refined process for doing so that I hope to test out. But yes, working in smaller steps is definitely a requirement, as well as constantly reminding AI of how we want to do things, packages we are already leveraging, etc.
Hopefully I find the right mix and can actually speed up rather than get bogged down by AI going in circles like it has been for a lot of my videos.
@paul-towers ah ok it sounds like you are in the same position as me. I am so much more productive with ai but honestly getting drained by the constant massive swing between speaking to a hyper intelligent god and a drunken toddler 😂
Haha that's a perfect description of how it operates
Please provide all the videos in the playlist Paul some of the videos are missing
Sorry, yer I'll update that