Jerry Brito

Threshold and organic generative art

Mon, 23 Feb 2026 14:02:02 +0000

Threshold and organic generative art

Notes for the week of February 16, 2026

Took Penny to the orthodontist and unsurprisingly she will be getting braces. She is sanguine to interested with the prospect. That surely won’t last.
Spent a lot of my time working on the feed reader / article discovery app I started last week.
- This week I came across a new reader app called Current that is addressing a lot of the same things I am. Ultimately it doesn’t work for me because I’m interested in training my system on what kind of articles I find interesting and, more importantly, what kind it should never show me, and this app doesn’t do that. But, it has a great design pattern I immediately stole: The user sorts the feeds into categories based on expected signal-to-noise ration (this is a breaking news feed, this is a personal blog, this is the New Criterion, etc.), and then the system applies different decay functions to each kind of feed. Articles from newsy feeds go away after a couple of hours, journal articles will hang around for a few days, etc.
- Now, when I open the app (which I call Threshold BTW)I see a list of new articles in my feed ranked by an expected interestingness score derived from thumbs up/down I give to articles. There is no unread count. Old newsy articles, even if I would have found them interesting are gone even if I haven’t seen them. I peruse articles from top to bottom as far as I want to go, marking some with a thumbs up (which puts them in my read later system), some with a thumbs down, and most with nothing. Articles from higher signal-to-noise feeds hang around longer and have a greater chance of being seen even if their expected interestingness is middling.
- That’s how discovery works. One the consumption side I use Readwise Reader and I simply sort my articles by time added. I just work my way down my queue whenever I feel like it. Their text-to-speech feature is great, though I wish they would add some of the newer unreal voices available. If there’s an article I come across that I think is worth sharing after I’ve read it, it’s pretty frictionless for me right there to mark it with a star and add a little comment if I want to. Those now show up in a Linklog section of the website I’ve now added.

- Also spent a fair bit of time on generative art and had a bit of a breakthrough. - I’d been coding up and plotting a number of pieces but I was not super happy with any of them. - First of all, I’m not crazy about using these tools for figurative art. The portraits and such that I’ve seen (and that I’ve plotted myself) are cool, but ultimately a bit gimmicky. - I imagine most folks feel this way because most of the generative art you see plotted are very geometric and structured. That’s the other kind of drawing I’ve made. I like these much more, but I have yet to really find them satisfying. They look more like craft than art, and I’m not really expressing anything. - The breakthrough I had was realizing I shouldn’t be looking at other pen plots for inspiration. I should just look at art that I like. And doing that, I came to the conclusion there’s not reason one can’t make more organic shapes using these tools even if I don’t see many people doing so, so that’s the direction I’m exploring and I’m pretty excited with what I’m cooking up. Maybe I’ll have something to show next week. - I’ve also continued to noodle on a more representation project that maybe veers from art a fair bit into dataviz. Basically I’m taking football statistical data and trying to figure out how to represent individual matches in an aesthetically pleasing form. What I’m coming up with reminds me of NFT drops.

Thank you for reading. You can view past weeknotes on my site.

Pen plotting and filling niches

Mon, 16 Feb 2026 14:02:01 +0000

Pen plotting and filling niches

Notes for the week of February 9, 2026

This week I continued my explorations of generative art and AI-related automation. These are the things obsessing me at the moment.
On the art front I made and released a new album.
- This one uses a different audio engine than the previous ones.
I’ve also begun making generative visual art.
- For a long while I’ve been interested in pen plotting, following various artists and communities online. I’d been hesitant to plop down cash for an expensive plotter, but recently came across a cheap crafting vinyl cutter that could double as a pen plotter—something that was marketed as an afterthought.
- Well, it arrived in the mail and I have been having a blast plotting. Most of the work is in coming up with the visual designs on the computer. There are a number of great tools online that one can use, but now I can also use Claude to help me develop generative tools of my own.
- Again, the artwork is not AI generated. What AI helps build is the software that algorithmically produces the art pseudorandomly within user-specified parameters. The artistic vision is human but the output is machine made and unique. I love it.
I’ve also spent a bunch of time playing with AI and automation. I got myself a VPS with n8n and moved over a lot of the personal assistant functions there.
- I made an automation that will post my photos to Instagram — something that I stopped doing because I hated the manual aspect of it. More interestingly, before a photo is posted, Google Gemini analyzes it and determines the hashtags that describe it.
- I’ve also made a number of automations that simplify my email, school scheduling, and other quotidian matters that required a lot of manual checking and faffing about.
And today, Sunday, I spent a couple of hours with Claude and coded up an app for myself that would previously have taken me months, if I would have done it at all. Basically it is a feed reeder that will learn from my tastes to increase the signal to noise.
- I can subscribe to high volume feeds and then I can give articles thumbs up or down, though it’s only a minority that get the thumbs up.down
- The aha for me came in conversation with Claude when I realized that I didn’t so much care about teaching the system what I found interesting so that it would bubble up those kinds of articles, but that I want it to learn what I absolutely will never find interesting and never want to see (the ragebait, the horserace, the culture war, etc.).
- Each day the system fetches the new articles from my feeds. It then has Ollama 3.2, running locally on my Mac, extract a list of ‘features’ for each article. Features are descriptive keywords for the article, but also source, author, and the like. It saves this along with the article in a database and these are all invisible to me.
- I can then go thru and read all the headlines and descriptions. For most items I can just read the headline to get the gist and move on. For some, I’ll hit thumbs up and that will give it a positive signal that’s recorded in the database and, simultaneously, the full text of the article is added to my Readwise Reader so I can read it later. And when I come across the kind of article I never want to see, I hit thumbs down, which is also noted in the DB.
- With these signals recorded, next time I load articles they will be sorted based on the predicted score based on the features it has compared to the scores for the features of the articles I’ve voted on. If everything goes according to plan, after a couple of days of doing this, all the crap articles will be at the very bottom of the list, which I might not even get to, and all the potentially interesting stuff at the top, but for this to be successful it only has to be right about the bad stuff. After a while, once I understand the thresholds, I could even institute a floor below which I won’t see items.
- If this works I would be able to subscribe to the full feeds of, say, the New York Times or The New Yorker or The Atlantic, without the mental friction that keeps me from doing so now: having to wade through all the garbage to get to the potentially really good stuff that is in there.
What all this stuff I’ve been doing lately is making me realize is that all the niches are going to be filled. The economies of scale of software production were still such that lots of little bespoke or niche software never got made — not even for oneself. But with AI that all goes away. This is going to have knock-on effects for the more-general-purpose producers for whose products one often had to settle. This guys makes the case that the app subscription model is going to die as a result. I can see that.
R.I.P. David Farber.

Thank you for reading. You can view past weeknotes on my site.

Generative music and Paco is born

Mon, 09 Feb 2026 14:02:03 +0000

Generative music and Paco is born

Notes for the week of February 2, 2026

Last week I mentioned I had begun playing around with generative music. Well, this week I released my debut album. 😉 Then I followed it up with two more. You can find my music on Spotify, Apple Music, and all the other streaming services.
- This is not AI-generated music, though it is created algorithmically. Basically I’ve developed a script that will create pseudorandom albums within parameters set by me. All the tracks have the same sound and vibe, but they are all pseudorandomly different.
  - This is something you could do before AI just with Python scripts, which is why I call it pre-AI slop. 🙃
- I initially developed it with an engine that generated Lynchian dark ambient music. I now have developed a second engine that will produce a ritualistic tribal ambient. (I’ll release an album using that engine soon.) I plan to write many more engines and then randomly choose between them for album releases.
- Submitting the albums to the streaming services was surprisingly straightforward, and thought it took a week to get the first one approved, now I can seemingly add a new one each day.

After considering installing my own instance of OpenClaw, I decided against it given the cost and security concerns. Instead, I decided to build my on OpenClaw-inspired personal assistant. I call him Paco and it’s working very well.
- Basically I connected a Telegram bot with an instance of Claude Code running on a Mac mini at home. It’s whitelisted so that I’m the only one who can talk to it.
- I also created a couple Claude skills that lets Paco read and write to my notes in Bear and tasks in Things (it’s not as straightforward as using things like Obsidian and Todoist, but aesthetics matter to me).
- I also created an email account for Paco and made a whitelist of who it can send/receive emails to/from.
- Finally, there are a bunch of helper scripts that run via cron for different functions, and it has a ‘heartbeat’ script that runs every 20 minutes with prompts for Claude to be proactive on several fronts. That means that every 20 minutes Paco will ask himself if there’s anything he needs to tell me or do for me.
- From my end, the experience is that once in a while Paco will message me to nag me about a task I haven’t finished and on which he knows I want him to follow up. I can also ask him questions about anything and he’ll use not just his general knowledge to answer, but also my notes and tasks, as well as persistent short-term and long-term memory.
- An aha! moment came when KO asked me in passing, ‘Hey can you send me the flight info for the Austin trip when you get a chance?’ Normally I would go into my notes, find the appropriate note, copy the info, and paste it in a message to KO. This time I just opened Telegram and spoke a message: ‘Paco, can you send the flight details for the Austin trip to KO?’ One minute later she had a courteous email from Paco relaying the info on my behalf. 🤯
- I’ve just added automatic package tracking and I plan to add SMS support soon so he can send and receive texts in addition to email.
Went to Penny’s second-grade play. She played the part of Balu in The Jungle Book—and she did it in front of an audience of almost 400. I’m very proud of her, and I’m also so impressed with her school. Each year every student has a speaking part in their class play, which they put on in front of the whole school plus parents and guests. I can’t remember the first time I had to get up to speak in front of 400 people, but it wasn’t in second grade.

Shoveled out my car, which was trapped in six inches of frozen snow-ice. At one point I got it stuck in such a way that it would have blocked KO’s car, too. Luckily I didn’t leave us carless.
Bees won away at Newcastle, which we hadn’t done since 1934!

Thank you for reading. You can view past weeknotes on my site.

Snow, Montaigne, Pyrrho, Schopy, and pre-AI slop

Mon, 02 Feb 2026 14:02:01 +0000

Snow, Montaigne, Pyrrho, Schopy, and pre-AI slop

Notes for the week of January 26, 2026

We got six or seven inches of snow on Sunday. The killer, though, is that it hasn’t gone above freezing all week, so it quickly turned to solid ice and the plows haven’t done a great job with that. School closed Monday, Tuesday, and Wednesday, delayed start on Thursday and Friday.
First meeting of a monthly reading group I’ve put together to get through Montaigne’s Essays over the course of a year. Great group of folks. Looking forward to seeing how this develops.
Recorded an episode of WAP about a collection of Arthur Schopenhauer’s late writings called On the Suffering of the World edited by Eugene Thacker.
- Stably was quite frustrated with it, while I loved it. That always makes for the best conversations.
- Discovered in the book what charitably can be called an easter egg or, alternatively, completely fabricated footnotes by Thacker.
Been reading Pyrrhonism: How The Ancient Greeks Reinvented Buddhism by Adrian Kuzminski. Puts certain Buddhist concepts in Western, rationalist terms, which has been very helpful for my very literal mind.
- Buddha ➵ Pyrrho ➵ Montaigne. Maybe.
Explored the idea of generative art further.
- Made a bot that algorithmically generates short videos. I programmatically distort images and produce a pseudorandom soundtrack for each. Once I tuned the look and feel I wanted, I could set the bot to upload as many as possible.
- Inspired by the “work” of Roel Van de Paar, I’m exploring whether quantity has its own quality. There may soon be a demand for pre-AI slop. In a couple days I’ve gotten over 12k views.
- Similarly I played around with a live-streamed generative numbers station.
- Also worked on making non-AI generative music. More on that TK.

Built a simple weeknotes blogging engine, the output of which you are reading.
Came across a guy making the case that Apple, in the long run, is best positioned to ‘win’ the AI race. Don’t know about all his reasoning, but I’m sympathetic to the argument.
- The more I learn about the business of AI, the more I find the current datacenter model unsustainable. OpenAI, for example, has entered into commitments to build over $1.3 trillion in computing infrastructure over the coming years. It has revenue in the $20 billion range.
- I am no expert, and I understand that the bleeding edge of frontier models need insane chips and compute, etc. But to the extent that you can give 80 percent of the utility of consumer AI to 80 percent of people with an LLM running locally, then Apple with its Apple Silicon chips might be sitting pretty.
A friend sent me this great article about the homogenization of downwardly mobile American youth culture. (Made me finally subscribe to The New Atlantis.) Seems a lot like Japanese low-desire society. Japan is the future.
What’s funny about the Moltbook story is that it is basically the plot of Her. It’s too on the nose. Feels like a prank to me.
Switched to Kagi for search. It seems ridiculous to pay for search, but it’s such a better experience. And a subscription comes with access to a suite of LLMs, including Claude Opus 4.5 if you pay enough.
- Digging their custom ‘lens’ feature where you create a set of site and you’ll get back results only from those.
Bees beat Villa away for the first time since the 1930s—and playing with ten men for an hour. Very proud of the boy (except for Schade with whom I’ve lost all patience).

Web version: https://jerrybrito.com/weeknotes/?note=05-2026

Update for week 43 of 2025

Sun, 26 Oct 2025 08:00:04 +0000

Photos I have recently added to my online portfolio.

Miami, FL

Pittsburgh, PA

Update for week 42 of 2025

Sun, 19 Oct 2025 08:00:03 +0000

There is a new episode of Worker & Parasite, my book review podcast with Stably. In it we discuss The Invention of Good and Evil by Hanno Sauer.

Photos I have recently added to my online portfolio.

Potomac, MD

New York, NY

Washington, DC

Update for week 41 of 2025

Sun, 12 Oct 2025 08:00:04 +0000

Photos I have recently added to my online portfolio.

New York, NY

Update for week 40 of 2025

Sun, 05 Oct 2025 08:00:03 +0000

Photos I have recently added to my online portfolio.

New York, NY

Update for week 39 of 2025

Sun, 28 Sep 2025 08:00:03 +0000

Photos I have recently added to my online portfolio.

New York, NY

Potomac, MD

Bedford, PA

Sideling Hill, MD

Pittsburgh, PA

Update for week 38 of 2025

Sun, 21 Sep 2025 08:00:04 +0000

There is a new episode of Worker & Parasite, my book review podcast with Stably. In it we discuss Total Defense by Andrew Preston.

Photos I have recently added to my online portfolio.

Pittsburgh, PA