The Exchanges

Q Interesting. I never thought about that, but that makes a lot of sense. Can you show us how to, are you going to show us how to do it?

A Yeah, so in this specific case, I would go to O-one model. I would just paste in the whole, like, doc we created so far, and this kind of a standard prompt I normally would use, ah, so I, firstly, I wanted to generate the file structure. Um, I was just pasting, and then we will see. So I, you can see that I'm pasting the whole doc, and in the end, I will ask it to, like, above the project, I want to build, how should I structure my project file? And that will give it a bunch of, like, common rules, which is very similar to what I put in here in terms of project setup. It's basically giving a bit more instruction, like, All new components should go to this folder. Page should go to this. We are using Next.js as this one. Uh, then based on that, uh, one model where first they try to propose, okay, this is like the final file, uh, like project structure or different components that we will need. Uh, so it know what, what towards, and you can probably take a look here to see if this align always what you were thinking, because sometimes you will identify some kind of very clear mistake. They might have repeat component four or twice, things like that. Um, but If this looks good, um, this, the benefit of this one you can see is that it actually will speed up, uh, do a bit more planning and architect the project a little bit more. Uh, and they try to also identify what kind of depend…

AI assessment note: “Yeah, so in this specific case, I would go to O-one model.”

Q I've never seen someone say, let's think step by step. Why did you, why did you write that?

A Uh, that's because, uh, like non, in previous large range model, uh, there's a very common technique called chain of salt. Uh, so basically once you get, uh, you kind of get it to switch gears. Um, so when they try to, uh, it's like system one and system two syncing, uh, in the, in the book kind of syncs fast and slow. When you try to get it to do just everything at once, then it often fail. Uh, but when it breaks down into small steps, it tend to have better result. And that's what we kind of try to simulate here. Uh, so once it tries to sync step, step by step, instead of giving you that two plus two equal to four, it will try to sync. Okay. How do I break it down into different steps? So that often will lead to better results. That's why we're kind of trying to prompt it for some kind of harder problems. Ah, ok, so this time, I can see that it gathered results. Um, and if I just try and change this to be, let's say, I'm happy with my current project, ah, but want to, ah, see how can I get my doc happier. And if I try this, hopefully that Uh, ok, so this one is not correct, uh, but happier, uh, and I'm looking for.

AI assessment note: “when it breaks down into small steps, it tend to have better result”

Q And when you get bugs, do you, do you get frustrated? Like, do you get a bad feeling in your stomach? How should one feel when, when you get bugs?

A Uh, uh, it, I think, yeah, that's why I, I definitely feel, didn't feel good. Uh, and hopefully some of them will go away later, uh, with, uh, with the newer and newer model. Um, but Yeah, it's hard, and, and that's why initially I want to kind of break that down. I learn, kind of, learn a lesson, like, you just keep getting in circle for even very simple problem when you're looking to the actual fix. Um, yeah, that's, that's, that's very frustrating. Um, yeah, like, for example, in this case, uh, it made a change, but you can see that clearly the change here is wrong. Um, So I were probably just to, but, but Cursor actually integrated a pretty good feature recently. I don't know if you saw that. Check it out. And this will allow you to kind of review that a bit better. So I can, this time I'm going to give it a bit more instruction. So every time when that situation happened, like I will kind of try different ways to either prompt it, like get it, think step by step, or give it a bit more instruction or additional documentation. So I got arrow above. Uh, please help me fix it based on the code example here. So we're, um, just copy paste again. Help me fix it.

AI assessment note: “I definitely feel, didn't feel good... that's, that's, that's very frustrating.”

Q While this is happening, it's crazy because you're a product manager. Now you're an engineer, right? So it's like, what is the role of the product manager if You know, because the old way of becoming a product manager was you'd write specs, hand the specs over, see product, and now it's you write specs and you build product.

A Yeah, exactly. I kind of think like this AI coding start, uh, enable everyone become kind of part of manager. Um, and that's really, really cool because I saw what's really unlock is a bunch of people who are prosumers that don't do this code now, but you suddenly have this ability. Um, Okay. Perfect. Now you can see that it returned the result of the Reddit post. So now we know this code example is going to work. Um, then we can, um, paste this in to the doc. So here I will say, uh, document, documentation for how to fetch, uh, Reddit posts. Using, uh, snow, uh, crap. Um, and I can say code example. And one thing I probably would also do is like, uh, try to reference a little bit. Um, like in this title, I will put for number three. So it's very clear. What's this for? Ok, so we get this part done, and the next part is another code example, probably for OpenAI. Uh, and OpenAI is like a perfect example. I guarantee the first time it will fail, because it is using some old documentation. Um, so we want to probably use structure output, uh, to categorize the posts. Um, so I can basically do something similar. So for this, I can add a new composer. I will add doc opening a structure output. Ah, so this time I will say, ah, probably just paste something in. Help me build a simple, ah, type script. Uh, script using OpenAI structured output to categorize subreddit for each category h…

AI assessment note: “enable everyone become kind of part of manager”