I’m acutely conscious, though, of each time I press the “I believe” button.

Maybe I should explain that term. When I was in the US Navy’s nuclear power training program, a quarter-century ago, we young sailors would have questions from time to time. The teachers were excellent and would do their best to explain, but at some point, if the confusion wasn’t resolved, they would have to cut it off in the interest of making forward progress. At that time, you might hear the beleagered petty officer say, “look, just press the ‘I believe’ button.” And generally, their judgment was pretty good - either it didn’t matter or or it would click. Either way, it was fine.

But pressing that button is an uncomfortable thing for me. Some things never click for me, and some things click but I forget them. As I continue working through improving my mathematical skill, getting into higher math, I find myself running into a lot of topics where I think, “you know, I never did really understand that concept…” I saw diagrams and graphs, I solved problems, I took tests, I got grades - but I didn’t capture its essential nature.

I don’t intend on letting these things just slip away from me anymore.

So I’ve added a number of additional notebooks to my queue to complete alongside ARENA - more on Einops, Einsum, PyTorch, NumPy, etc. You should see them start popping up on BottomUp Labs, probably in a new folder.

So that makes 7 of 98 parts, averaging 4.2 a week, which projects a finish around Apr 11, 2027. Which is still substantially later than I’d like. Aside from health issues, I have been going faster, so I’m still hoping for a finish in February if not earlier. Not that I’m trying to rush through or anything.

I came back today by… largely repeating the exercises I had done already, but in the 0.0 Jupyter notebook instead of the .py file. The .py file is recommended for the in-person sessions, but doing them in VS Code is not incredibly ergonomic on my laptop, so I switched to the Jupyter notebook.

The Einops stuff is pretty easy for me at this point; I had created some targeted Anki cards and the operations come to me pretty quickly now 🎉 I think that’s a sign that I’m on the right track with my study practice (for the first time in my life).

I ended up getting to E - Classifier Accuracy, but I started feeling quite shaky about things. I’m making some cards to build up my skill with the atoms leading up to those operations and combinations thereof.

Currently, I think this is scheduled to take me until April or something - I really hope that isn’t the case. I think now that I have my study systems in place, things should go a lot more quickly.

But we’ll see.

Today (11/05) I completed What is torch.nn really?, which is an introduction to PyTorch’s neural network modules, classes, etc. Embarrassing to admit, but I haven’t really worked much with it before, just some tinkering. It was really neat to see how simply models can be trained, modified, etc, especially with switching from a simple neural network to a convolutional neural network.

I’m also working through the first notebook of a series I’m working on about interpretability, using NES games: Meet Mario. It’s not finished, and I ran out of time today, but it’s coming!

I went through the tutorial today - or tried to, while watching the election results - and it was a bit of an overwhelming experience. Part of that is that this has already been an incredible day.

But more than that, reading through the tutorial, I realized that I haven’t been taking advantage of the incredible tools that we have. I’ve missed out on so much over the years by not exploring Python!

Today (11/04), I worked through the Einops basics again, and then Einsum is all you need. The former was great fun! The latter took a good amount of thought to understand what was going on at first, but I feel like it’s sinking in a bit. It’s shocking to me how beautiful and powerful things like this can be.

Today, I reviewed the planner and appreciated the full depth and scale of the ARENA program and how much I would have to work each day to complete it in a reasonable timeframe².

And then I reviewed 0.0, or Prerequisites.

Before going further, I think it’s probably worth explaining a bit of my background, especially with regard to mathematics. I graduated high school with a cumulative 1.667 (D+) GPA³. I unfortunately have what I refer to as “weapons-grade” AuDHD⁴, which went undiagnosed until I was 36 and untreated until I was 44. It was essentially impossible for me to focus or effectively prioritize work. I failed most of my classes throughout middle and high school, and I think I was graduated largely so they wouldn’t have to figure out something else to do with me. This was not conducive to a meaningful mathematical education 🙃

From the time I was a child, I wanted to be a scientist and to work with computers. But I thought I was stupid, and so I ruled those possibilities out until I was well into adulthood.

The US Navy prized my ASVAB score over my high school transcript, and so I enlisted and entered into the nuclear power training pipeline. My service isn’t terribly relevant except for that it’s how I managed to go back to college and graduate, at 31, with a degree in Computer Science, and why I am now working as a contractor for the Department of Veterans Affairs, trying to improve access to care and benefits for other veterans.

But my mathematical skills never really developed. I struggled to pass Calculus, and I graduated having felt like I just scraped by.

In August of 2025, I started using MathAcademy. I started with integrated math and am now roughly at the point where I was when I graduated college - linear algebra, differential and integral calculus, and probability and statistics.

So this is a long way of saying that I’m approaching the prerequisites for ARENA with some caution.

After some thought, I’ve decided to spend this week (11/02-11/06) focusing on some conceptual and programming prerequisites:

• Neural Networks: 3Blue1Brown chapters 1–4
• Python:
  • Einops: Basics Tutorial and some exercises
  • Einsum: Einsum Is All You Need and some exercises
  • PyTorch: What Is torch.nn Really?

I’ve also added the rest of the material where I feel weak or underdeveloped (which is a lot) as work to do alongside ARENA where most appropriate.

First up, I have some videos to watch 🙂