AI: A Concern Of Mine

March 20, 2024

AI is a subject that I was always ignorant about, kind of; I just didn't give too many fucks about any of it. It always seemed like a stupid thing that regurgitated human-readable data in a stupid manner. But lately, I have become more conscious about the subject, especially after the announcement of Devin, the AI that can do the job of a software engineer start to finish.

ChatGPT

Let's keep the discussion about Devin until later. My first interaction with AI was through ChatGPT when it was first announced in 2022-2023. I wasn't impressed or curious about it for a long period of time when all my peers in my school were using it all the time. I just wasn't interested because I brushed it off as another stupid new overhyped thing, but it wasn't. ChatGPT became a tool that I used daily to complete a lot of my tasks, many of which were programming-related. It fully replaced my constant need for googling and browsing the web for information as it provided most information I need on any subject in a clear concise formula. It also helped a great deal with tedious/repetitive tasks such as creating mockup data. Much later, I also started using it to write letters and improve the grammar and styling of my writings. It overall was just a really useful/helpful all-in-one tool for assistance given by separate internet tools.

While extremely useful, ChatGPT was and is still extremely stupid; it's still not smart at all in my opinion, as it can literally not do anything but regurgitate information it has already seen. It writes HORRIBLE code most of the time, especially for things like HTML and CSS which obviously based on the data it was given would suck no matter what. Almost nobody out there writes good HTML or CSS; things like ReactJS and Python are written in a much slicker manner because they're so readily available wherever you go.

Devin

Then Devin comes into play. First things first, I have to preface this by saying that I'm extremely skeptical about the claims in the announcement video because they're just claims; nothing's available for public testing. The data sets on which they spit out the 14% success rate on unassisted prompts are nowhere to be found; nothing is credible. It was just a cool video and that's where it starts and stops, but this whole thing brought to my attention that the leaders of this AI wave are OpenAI and...? Oh fuck, where are the Googles, the Microsofts, and the Amazons? What the fuck are the companies with all the data on all humans on earth and all the money to buy as many GPU warehouses to train models with as much horsepower as there's energy available? It cost OpenAI $700M to train their language model; that's expensive as fuck, and it's quite reassuring, but only when you look at it from the point of view of most investors. If you think about the monopolies, only then you realize how fucked it is that none of them are announcing/talking or doing anything about this AI wave.

Copilot

Another thing that is quite scary to me is just how good GitHub Copilot is. It also became a huge part of my programming journey, as I think I'm already at a good enough level that I can allow as heavy of an assistance because I'm way past syntax errors and knowledge of how programming works. GitHub Copilot has all written code in history available for it to be trained on; it's on GitHub, so it's only normal that it's just so good at predicting what you need to write next. It also is really good at adapting to your own codebase, something that I found extremely useful, but there's something to note here, because the training data set is so huge and precise, the model is almost too good, and that just shows how much the variables can scale the 'intelligence' of the model off the scale.

Conclusion

In the midst of all of this, I see most people brushing it off with no valid arguments or any critical insight. It does look to me like we're nearing a huge change once again, in how humans are gonna live. I almost feel like a post-apocalyptic reality is upon us; people are yet to realize the endless possibilities that are coming. Software engineering seems to be getting all the attention, but it's only due to the job being so computer-focused, making creating interfaces through which an AI can interact with a human job much easier to make, but that's only a bubble that will soon explode.