Austin Steady

New year’s resolution for me was to post something once a month. Doesn’t matter what, ideally something meaningful / useful / tangible, but whatever.

The theme of January for me was a lot of agent-fueled ADHD. With the improvements in agent coding, my “productivity” has gone up a lot. I redid some old side projects and worked on several new ideas. It’s really amazing how something that took me multiple weekends and nights can be banged out in an hour or less now with zero manual coding. However, most of the new projects are still totally halfbaked. But what should I expect from almost zero effort? I imagine in a year from now, these will be 80% baked if not…

I live by a cliffed beach, but due to erosion, it’s only rocks most of the time. At low tide though, you get a wide stretch of flat sand perfect for walking.

The caveat is that not all low tides are low enough, in fact most tides are too high. The cobblestone part of the beach is steep and permeable. This means the tide can fall several feet from high tide without much noticeable visible change. You would still only see cobblestone and it would seem like the water is coming just as close to the cliffs as before because the vertical change in the beach is much greater than the horizontal change. However, once the tide reaches low enough…

I published a book! It’s called The Physiology of Climbing, and you can buy it on amazon or on my publisher’s website sharpendbooks.com.

I started the research for the book in June 2022 and finished the first draft in February 2023. After a few tweaks I submitted it for publication. Life got busy for both me and my publisher, but the book eventually got published in February 2025.

Following up on the exploratory work I did for searching recipes with OpenAI. I quickly found this blog that benchmarked OpenAI’s results against other, open-source, alternatives. Turns out you can do this stuff for much cheaper than I’d believed.

I’ve been taking a lot of notes on books and papers I’ve been reading about climbing physiology. I’ve got a Google doc that’s about 120 pages now of bullet points. I lose info in it a bit because docs only supports ctrl-f / exact keyword search. I got into the habit of using specific terms when reading about a specific theme like “collagen synthesis” as a shoddy tagging system.

I recently got access to OpenAI’s beta, and I’ve been playing around with it. It’s not perfect, and at times I wondered if I was just reinventing the wheel with the stuff I was building. But it is really cool. I do think it’ll transform our jobs quite soon.

(Code available here)

The idea behind cryptographic communication is that it should be hard/impossible for eavesdroppers to decrypt the messages, but relatively easy for the intended recipient to decrypt.

For this to work at all in general, the recipient must know something that eavesdroppers don’t.

At a previous job, I had the pleasure of improving our web app’s performance. The motivation was that we had recently experienced a number of outages in a short timespan, after having had no outages at all in the year I had been working there at that point. We were able to observe that our database cpu utilization was maxed out. Using AWS RDS Performance Insights (PI), we were able to see that a few queries were extremely non-performant.

PI shows you the average number of rows scanned per query, so it was easy to see that this wasn’t strictly a problem of increased traffic; we knew upfront that our queries were bad.

I’ve written a few apps that run either all the time or are supposed to run a few times a day. They live on my laptop until I get sick of the unreliability, and put them on a remote box. Unfortunately, I take some liberties and shortcuts with my personal projects that end up making the deployment more annoying than it should be. Absolute filepaths, relative imports, local db, etc. So dockerizing and deploying isn’t a mindless, trivial exercise for me. I’m hoping to basically make this a checklist/cheatsheet for myself to make the process faster in the future.

Build basic python image. Install necessary things like sqlite if you’re using it. Setup base directory (change the name). Install pipenv.…

The goal of this post is to learn about Markov Chain Monte Carlo. What is it, why does it exist, how do we “do it”, and how can we apply it to a real problem?

Markov Chain Monte Carlo (MCMC) sounds cool. Maybe you know what Markov chains are and/or maybe you know what Monte Carlo methods are. I was familiar with both, but I had no idea what MCMC is.

I recently realized that I didn’t understand linear regression as well as I thought as I did when I came across this twitter thread.

Okay, I thought, so maybe I don’t know exactly how to do it with linear algebra but I intuitively know how to draw a best fit for a set of points, so I should be able to work from there and develop a more rigorous algorithm of what I’m doing by intuition.

The market’s been volatile recently because of the COVID-19 pandemic. I’ve been checking individual stocks manually each day for large movements. I don’t want to daytrade, but I do want to have a general feel for what the market’s doing each day. But checking manually is annoying and also not healthy, so I automated the process:

Note:

I visited home for the holidays, and during that time I started making a surfboard. I didn’t get to finish before my flight back unfortunately, but I got pretty close. I need to apply a second hotcoat to fix up some contamination, and I need to install the plugs. But I think I learned a good deal and want to share some tips here.

I went to Greenlight Surf Supply for all the material. My dad luckily has pretty much every tool I would need for the process (power tools mostly just for building the stands, a random orbit sander for sanding the laps, levels and l-squares). In addition to the materials, I picked up one of Greenlight’s rail runners…

I came across this theorem recently, and thought that it’s an interesting result. The proof itself isn’t particularly flashy, but I think the result is relevant to every prorammer’s daily life and that’s pretty cool. My hope is to explain it enough that a programmer with basic logic experience can understand the proof. I’ll explain notation along the way.

Theorem:

This post is intended to help those with Popliteal Artery Entrapment Syndrome (PAES) by recounting my personal experience with the symptoms, diagnosis, surgery, and results.

I also hope that people who are developing PAES symptoms without knowing why might find this post from googling and be able to intervene early. My symptoms were local to my lower right leg and foot: pain when walking even small distances, increased sensitivity to cold, numbness, discoloration, Raynaud’s symptoms, pale skin and blue nails, calf soreness and sharp foot pain when exercising.

I had some issues getting rustfmt to work in spacemacs on my mac. Here’s roughly how I got it working.

These two github issues helped a lot: 1 , 2.

Detecting blurriness in a picture is useful for determining image quality since people rarely ever consider a blurry picture to be a good one.

We’re going to use the Fourier transform to easily analyze the overall spectrogram of a given picture. We’ll do this for several pictures, manually mark them as blurry or not (the method itself will ‘rate’ the blurriness, we’re just creating a cut-off point), and then feed the transformed images to a very simple model that will learn to classify the pictures as blurry or not.