AIblog1

AI is getting big in the coding world, and I can easily say I don’t write pure syntax anymore. Advanced models such as Opus 4.7 know C++ syntax so well you can almost guarantee that, given some parameters, it will not make syntax mistakes. If you specifically instruct your coding AI to avoid new, complex, or rarely used syntax and stick to what works now, has worked since the start, and will work years from now, it will not make mistakes.

The only mistakes AI does, excluding rare brain explosions like deciding to wipe a production database or similar that don’t stem from AI mistakes but more from AI overhead and instability, the only mistakes it makes at coding are logic, like not applying a math formula in the correct way. Such as, say, building a 3D mesh, it might flip faces or cull the wrong side. And the other being context-based, like forgetting that something needs to be changed in 2 places, maybe in 2 different files or such.

And the obvious being when AI is used unguided in a task that doesn’t fit into the “general AI knowledge” or hard-to-research stuff. You can build an entire operating system using pure AI, but you can’t go “build me an operating system.” You would have to understand all the non-syntax parts of an OS.

Say we live in a world where an AI is not trained to know that 2 + 2 is 4. AI can still be used to code a solution to the problem, given you write English syntax like in math, “2 + 2 = 4, code this in Python” or similar. This is called abstraction.

Back in WW2, solving such problems using a computer would take programming a set of dials with 0 manuals and 0 abstraction. Then you could sit with buttons inputting 0 and 1. Eventually you could write code in a program. Now you speak or write in your language, and the AI abstracts that to code, and the compiler abstracts to OS instructions, and the OS abstracts to interact with your machine.

Many believe this removed coding jobs, and maybe they’re right, but more I believe this shifts the job title. At the start, a programmer was 99.9% knowing how the machine and syntax worked. Now it’s 99.9% knowing the product you’re making, the systems you work with, and how to build on abstraction.

I don’t believe a new level of abstraction is needed. Abstraction is used to make our lives developing easier, so the next step after 99.9% abstraction would be auto development.

My verdict is that, given 10 years, the job is knowing what direction of technology the human race should research. Imagine sitting at a terminal, directing a set of autonomous AIs on what to explore.

The main thing is that, given the right I/O, AI can do anything. Currently almost all they have is text in and text out. That unlocks most computer tasks. Give them, say, a robot arm, they can do anything. And yes, that would be dangerous, so what between?

Give them, say, 10 different machines to research and explore. Say a fabricator of some kind, that kind of stuff. Then create hard-programmed use cases of these machines that act as guidelines, given that the AI can’t do shit other than the abilities given... as long as it doesn’t, idk, break out of containment...

Yea, let’s not give Mythos a fabricator. Maybe a less capable model, but Mythos... personally I’m thinking we just burn the thing right now and try again. Once we know 99.9% that we can control the AI, then try to make something like Mythos.

But right now we’re making nukes while being unsure if the fuse is working... doesn’t sound so smart, does it?