Whereas previous major advances in computer science and its applications always made me feel empowered, AI felt different.
In order to find out why I feel this way, I have to look back over the past few decades and compare the major technological advancements with today's technology.
PCs are the entry point for people to enter the world of technology – they get people to start playing games and eventually programming. I have fond memories of using tools like memmaker to squeeze more memory out of my machine so I can run some pirated copies of new games on my DOS machine.
PCs come in different shapes, sizes, and architectures. They go by many names, by different merchants. You are basically free to do what you want with these machines. You can replace the parts. You can run games, you can write software, or do something completely different with them.
From an engineer's point of view, the inner workings of these machines are understandable. With a little skill and dedication, you can even build your own.
Suddenly, the bandwidth of an internet connection is more interesting than the clock frequency of a computer. People can spend hours chatting with strangers in internet cafes, and they, like me, are exploring this novelty**.
The early internet was convenient for people. It is built on open, interoperable protocols such as TCP IP, HTTP, SMTP, etc. You can buy a book or access selfhtml, open an account on Geocities, and create one**. If GeoCities fails, you can move your cities elsewhere or even host them yourself. No single company or entity owns the majority of the internet.
New form factors and interaction paradigms have been introduced. Although we've seen similarly sized devices and touchscreens before, with the help of ubiquitous mobile internet access, the iPhone was the first to bring all of this together.
In most cases, smartphones are accessed in the same way as other computers. They are small, interconnected, programmable computers with a variety of input sensors.
I understand how smartphones work, I can program them, and I know their limitations.
How artificial intelligence is different
Don't get me wrong, I'm excited about AI and shocked by something like the just announced sora text-to-**model. But at the same time, I felt like I was being left out.
I want to understand how things work. Artificial intelligence is like a black box to me. In order to really understand why a certain prompt x leads to a certain output y, I have to read a lot of ** and digest a lot of math, which overwhelms me. Even some of the top scientists in the field admit that we don't really understand how AI works.
"If we open ChatGPT or a similar system to look at the inner workings, you're just going to see millions of numbers flipping hundreds of times per second," says Sam Bowman, an AI scientist. "We just don't know what that means. ”Without understanding how something works, we're doomed to be just users.
On the surface, of course, this is true – anyone and their dog can open a ChatGPT session or throw some JSON in OpenAI's API. What I'm talking about is getting the core technology base that makes AI possible.
Flipping all these numbers to get results (inference), especially identifying them first (training), requires a lot of resources, data, and skills.
Artificial intelligence is not a shovel for the little people.
If you're like me, you might have some ideas for building something cool with AI. But in doing so, you'll most likely end up building a GPT wrapper.
You may ask, what is a GPT wrapper? It's any software, SaaS, whatever you want to call it, that relies on someone else's AI product that you can't easily copy or swap (e.g. ChatGPT).
If I build an application that requires persistence, I'll probably use postgres and s3 to store the data. If these are no longer available, I'll use another relational database, key-value store, distributed file system, etc. But what if OpenAI decides to revoke access to the API features I'm using? What if they change the pricing and make it uneconomical to run? What if OpenAI expands their product and makes my product redundant?
None of this is known.
Artificial intelligence