alltom.com,
Archive,
Currently Reading,
Have Read,
Micro.blog
Subscribe with RSS or @tom@micro.alltom.com
What a delight. A new album from Dengue Fever. šµ
What I do is, I cancel every automatic update request, accidentally spot a news article about a āclicklessā 0-day exploit being patched in macOS, then I manually trigger an update.
My Steam Deck š® needed a charge and I couldn’t find its usual cable, so I plugged it into the USB-C hub that I use for my laptop⦠and the peripherals just worked. šÆ
My game appeared on my monitor, the audio played through the speakers, and my keyboard even worked.
Why had I never tried this!
When I write code that I know has noticeable limitations, I used to denote them with TODOs, but lately Iāve been writing them as disclaimers instead. Like instead of:
// TODO: Make this work for multiple cameras.
Iāll write:
// This only works for one camera.
I think itās a good thing to write only the code that I need right now, but noting the limitations with TODOs made it feel like a liability. It also probably gave the false impression to my teammates that I intended to add unneeded features that I really didnāt.
Imbroglioās current Izu mode was such a satisfying nut to crack. š
Currently reading: The Pathless Path by Paul Millerd š
Magna-tiles on a Saturday morning
Currently reading: A Memory Called Empire by Arkady Martine š
Who remembers Branch? š Maybe they still have the code around and can fill this Twitter-shaped void.
Yesterday, I wrestled with insecurity about a design Iād proposed at work. I played Anima on headphones and paced by my desk until I didnāt think the people working nearby could take any more, then I moved to an empty conference room. I think I believe itās the best option again?
Me, staring at an LLM answer: “Hmm. Was the training data that contributed to this answer based on sponsored content?”
ĀÆ\_(ć)_/ĀÆ
Introducing Chat Notebooks: Integrating LLMs into the Notebook Paradigm
What a time to be alive.
I think I’ve used more OpenAI credits in the last hour than I have over the last month. š
I just created a Google Doc from Mathematica. I feel unstoppable.
Today I wrote a quick little post about exploring JSON REST APIs in Wolfram Language.
I wouldn’t have thought that Mathematica would be such a great environment for manipulating JSON, but its built-in dataset query APIs and table renderer are just good!
People think that when they turn their eyes from the earth to the sky they see the heavens.
The One-Straw Revolution by Masanobu Fukuoka š
ChucK continues to shine. šµ Todayās find: a real-time audio mosaic that plays audio clips from another source (in this case, A Day in the Life by The Beatles) whose features match the audio from the microphone.
The latest draft of my Wolfram Language tensor math article is up! I get into function minimization (kind of important in ML?) and pack in several cool purpose-built visualizations that take only a few lines each.
I’ve already drafted the next few sections and there’s fun stuff coming up⦠š
Elmo can do no wrong⦠except the Happy Dance. My kids will not abide it. They yell, āNo Happy Dance!ā as soon as Elmo learns something new. I donāt know why.
I just posted a massive new update to my tensor programming article draft, with an expanded introduction, a worked neural network example, matrix equation pretty-printing, and several gorgeous plots! This article is proving exceptionally fun to write. :)
I have a big ol’ writing project planned for exploring fundamental ML concepts in Wolfram Language.
When I’m trying to get work done, I use JAX, but when I’m trying to understand and visualize, I prefer to work symbolically. This article is going to demonstrate why. :}
Look at how dramatically Wolfram is rebranding. In their own marketing email, they introduce the companyās primary product as āWolfram Languageāthe language behind the Wolfram plugin for ChatGPT.ā
My workflow is nascent, but this is the snippet that allowed me to drop that last Mathematica notebook into my site template with no modifications:
Export["body.html", EvaluationNotebook[], "HTML",
"CSS" -> None,
"FullDocument" -> False,
"GraphicsOutput" -> "PNG",
"ConversionRules" -> {
"Title" -> {"<h1>", "</h1>"},
"Section" -> {"<h2>", "</h2>"},
"Subsection" -> {"<h3>", "</h3>"},
"Text" -> {"<p>", "</p>"},
"Input" -> {"<pre>", "</pre>"},
"Item" -> {"<li>", "</li>"}
}]
I follow that up with cat head.html body.html tail.html > index.html
.
By default, Mathematica turns all those cell types into <p>
with classes like “Title” and “Section”, but once I switched to the semantic equivalents, they just worked.
I still haven’t figured out how I want to do margin notes, or how to wrap lists in <ol>
or <ul>
, or how to embed text versions of all the inputs that it renders as PNGābut in this case I felt that perfect was the enemy of good!
First they trained ChatGPT to answer questions with Wolfram Language, then they built a library for making your own functions available.
I’m excited about Mathematica again!
As a warmup for publishing more, I upgraded my contrastive representation learning notebook from PDF to beautiful HTML.
Iāve been noodling on one to the tune of āLet It Beā:
When my planetās spice is in trouble
Itās time to train a new army
Seeking words of killing
Like Muadādib
Like Muadādib
Like Muadādib
Like Muadādib
Like Muadādib
Seeking words of killing
Like Muadādib
These could be exactly the same oil and I donāt care. The containers are SO much fun.