introducing hundomentals

It is extremely tempting to completely ignore the importance of the fundamental, base-level knowledge now that a lot of stuff can be done for us. I’m not just talking about AI, either.

JavaScript frameworks made it insanely easy to ignore JavaScript and web fundamentals. Electron made it insanely easy to ignore GUI development in desktop environments. The browser made it insanely easy to ignore the fundamentals of memory management and operating system interactions. Game development engines like Unity and Unreal and Godot make it insanely easy to ignore how things are drawn to the screen, how memory is managed, how collisions work, etc.

Now that we have AI systems that are capable of abstracting our knowledge even further, it is insanely easy to not have any clue what’s going on “under the hood” as we write software, and our definition of “under the hood” is expanding daily. This doesn’t just apply to software, either. We run the risk of outsourcing all of our knowledge of the ways the universe works to AI.

This isn’t an anti-AI post. I’m not anti-AI. I like it a lot. The idea of this blog and the efforts that are going into it came from a conversation with Hermes and some journaling I’ve been doing since I’ve started using AI for a lot of things.

We cannot abandon the understanding of fundamentals.

In the post-LLM age, it’s more important than ever to use the technology that we have been given not to widen our pool at the detriment of depth, but to use the tools to dive deeper. As the libraries opened to the masses, we didn’t create more generalists, we allowed more and more people to become specialists. LLM’s should be viewed in the same way. They are flawed libraries with massive potential, potential we can harness to create experts in esoteric, low-level knowledge.

Codeslinging as a profession is dying, but software development and architecture is going through a renaissance period right under our noses. You will not be valued for the production of Lines of Code, but for your understanding of complex systems, how to debug them, how to make them move, how to wrangle them. In order to really understand these systems, it’s vital to understand them at a lower level than you thought, especially if you want to truly harness the velocity that AI can produce.

For example, what percentage of LLM users know how AI systems work? Not at the API level, not at the code harness level, at the transformer level. I sure as hell don’t, not yet. As time goes on, we need the number of people who understand these concepts to increase, not decrease.

As artificial intelligence threatens to increase the number of reported vulnerabilities in the wild, and as sloppily produced code threatens to create more and more vulnerabilities as well, we need more people to understand the core mechanics of security, not less.

As such, I’m doing some restructuring on this blog.

the hundomentals (100% of the fundamentals)

The hundomental tag and structure is what I’m going to use to build out guides for material in subjects I’m studying. The key purpose of these guides is to build a full guide from the most basic fundamental knowledge of a topic up to its higher levels.

Let’s take AI agents as an example. To understand AI agents, you have to understand:

• what is meant by the word agent, which requires knowledge of
• what a tool call is, and
• how tool calls work at the LLM level, which requires knowledge of
• how LLM’s work, which requires knowledge of
• how LLM’s are trained, which requires knowledge of …

So the goal here is to be able to go from the minutiae knowledge of the mathematics that go into transformers, all the way up through the way a model is trained and post-trained, all the way up to agent logic and how Hermes works.

This mode of learning (and teaching, which is a fantastic way to learn) means you can zoom in and out at any level and have an in-depth knowledge of the systems you’re talking about. Ideally, there’s close to zero level of abstraction that you cannot at least sort of grasp.

This is applicable to… just about everything. Even subjective subject matter, things like politics, have levels of fundamentals that you can zoom into and out of. I’ve written about Pete Hegseth’s Jerusalem Cross tattoos, which to properly understand you can go all the way back to the early Crusades, or to the tenets and aesthetics of Mussolini’s fascism, etc. If you want to understand the Israel-Palestine conflict? There is… a lot of core, fundamental reading you can do, going all the way back to the days of Abraham and Muhammad.

So, practically, what does this look like?

Over the next few days, I’m going to be restructuring the existing posts on the blog to have links to pages that contain “precursor information.” What precursor information is is “lower level”¹ information that adds context or understanding to the article at hand. So the precursor information for an article on Hermes might be “agent harnesses” so there will be a link to an article on agent harnesses, which might have a link to an article on “tool calling” as precursor information.

After each article will be another section for “higher reading.” This is “higher level” writing that would view the article at hand as either sibling (same level) or precursor information. In the beginning, most of these articles will 404, as I haven’t written them yet, but over time I’ll fill them out with deeper dives on the lower level stuff.

I have several purposes for this effort:

• enumerating blind spots - selfishly, this will help me enumerate areas that I don’t have a proper understanding of. I know how to create a React component, but I couldn’t tell you the first thing about how the rendering engine works. I can tell you that Pete Hegseth sucks, but I can’t tell you why the Crusader tattoos are actually problematic.
• encouraging others to take a fundamentals-focused approach to learning - at a higher level, I want more people to start digging deeper into their own understanding of topics. Yes you understand how to make REST API calls to Anthropics servers, but do you know how inference works? Yes, you understand that the liberals policies toward drug intervention have failed in California, but do you understand the history of policing and the US penal system?
• personal growth - as my understanding in these topics grow, my capabilities will as well, unlocking a lot of potential avenues for employment, research, startups, etc. I believe this type of focus, and the information that comes from it, can benefit others in the same way.
• research is dope, for its own sake - more knowledge is almost always good.
• gaps in LLM input/output - I’ve noticed that LLM’s in particular often lack knowledge of lower-level, fundamental stuff. They oftentimes don’t understand things like complex historical conflicts, Windows API esoteria, etc. Viewing content, especially public content, as eventual LLM output, this type of focus is helpful for LLM’s to get training data on lower level knowledge, especially if it catches on.
• this is the type of content that I believe helps cleanse the internet - on a deeper, more meta level, content like this I think is what can make the web a better place. Getting back to treating it like a library instead of a casino or a place to scam people out of money is a good thing. This is a very small corner of the internet, but it can make a difference.

organizing

For now, I’m going to use tags to organize things, but eventually I want to have an actual hierarchical layout. More to come on that. I’ll use hundomentals-<topic> to organize things into topics and have the pre-cursor and higher level learning articles as links. My goal is to not break any links in doing this, both because it’s annoying and it breaks SEO to do that.

Footnotes

1. When I say “lower level” I mean it in the ‘computer science’ sense: a lower level of abstraction, not a lower level of simplicity. So C would be a ‘lower level’ language than JavaScript since it runs on the operating system that runs the drivers that run the browser that runs the VM that runs JavaScript. Understanding Islamic history would be a ‘lower level’ than understanding the Israel-Palestine conflict, as most of what is going on now is happening in a broader historical context. This isn’t to say that nothing is running outside of this context (JS does things that C cannot adequately explain, Bibi’s corruption is happening roughly outside of the context of the history of the region) but it is, generally, at a lower level of abstraction ↩