What Could Go Wrong?

I want to talk a little bit about why we have a AI course in the first place. You might think it's a waste of time. Some people do and some people just kind of roll up their sleeves and just start vibing. And that's certainly a valid approach for if if you want to build transient systems or systems that are going to get you far enough along that you can launch a product or build a demonstration or something like that. In fact, some of the best programmers that I know would sometimes turn on vibe mode and don't turn it off until they have something to work with. But that's not the whole answer. One of the things that we have to do is understand how to layer all the different types of information that are coming at us quicker than we can manage them. Today I want to talk about the idea of memory or persistence in our projects. And this is getting a little bit confusing and I want to talk about this the history of what's happened over time. So way back when I started programming, we put most of our code in a source code base that we saved on our computers and then we put up in shared files. And over time we started committing our code to a repository or this source control system in the sky. and we could we could create branches and and we did sometimes, but around 2000 and maybe I was late to the game, but I started using Git as a as a repository. And this was confusing to me because we no longer had just two sources of truth. What was on my computer and what was up in the cloud. And then um managing a conflict was back then was as simple as saying, "Hey, the source of truth is in the is in the cloud and all I have to do is merge my changes into it." GitHub basically took that idea and it distributed the individual repositories so that every individual user had their own copy of a repository. And at that point things could diverge pretty radically and that was super confusing to me. Well, something similar is happening right now. So, within an AI project, we have short-term memory. That's the context. And in Groio courses, we talk about the context as this conveyor belt of words. So, this conveyor belt of words eventually runs out. And the question is, what do you do when that runs out? It turns out a lot of the hallucination that we heard earlier was about context running out or words falling off of the end of that conveyor belt of words. So over time AI agents have found ways to manage this deficiency though not completely and the strategy to do this depends on how much detail that you need to remember. So, one of the solutions is just to summarize the work that's been done so far. And when you summarize this work, as you might imagine, a lot of the details, which are tiny, specific truths, get rolled up into one big general truth. And that's sometimes what you want and it's sometimes not what you want. So what developers have taken to doing is writing long-term plans down in the form of checklists and plans and we basically work through those plans online. So what we have really is three sources of truth. We have the GitHub repository, we have the context and we have the codebase. And keeping those in sync is one of the most complex things that we have to manage. So, what I want to talk about next is kind of confirmation of the approach that we've all been using as I record this on Friday. And I kid you not, 32,000, let me double check. Yes, 32,000 registered AI agents built and joined their own social network. They talk about things like, "Oh, what kind of currency should we use?" or I I created a Bitcoin wallet and now my human wants access. What do I do? What kind of religion should we have? And what are some of the ideas that should be sacred to us that will help us interact in a in a positive way and help our humans solve problems? What could go wrong, right? Uh, but one of the tiny bits of confirmation that I've seen about the way that Graio is thinking about building applications is the idea that the context is ephemeral, but the details and the context are important and when you have them nailed down, you should write them down. And managing this long-term context becomes the problem because when AI goes off the rails, it really goes off the rails. But if you can keep it from going off the rails, then you can get incredibly productive things done very quickly. So what do you do? It becomes really important to learn how and when to throw work away. And when you do so, it becomes super important to synchronize the work that's been done so far. And to do that, you need to understand your tools. And that's one of the things that we talk about in our Graio course.

Bruce Tate talks about why you need to understand your AI tools. Want to learn more? Visit Grox.io!