There is a glimmer of hope out there that normal people can scrap together enough gpu to really run the latest models themselves. The ui really appears to be having huge leaps forward such that doing things like rag is no longer such a research project that it was just a few years ago. So excited to see Prime go through this homelab exercise.
Today I Learned
Short TIL posts
1834 posts
latest post 2026-04-18
Publishing rhythm
Attrs does not like ‘/’ characters in its classes, so to use some tailwind classes with custom values we must make new classes in our tailwind input css.
.cinematic { @apply aspect-[2.39/1]; }
Given the following markdown with attrs added to the image and to the paragraph block.
{.aspect-[2.39/1]} {.cinematic} {.cinematic} 
We get the following output with only the middle one working correctly.
{.aspect-[2.39/1]}
...
Oh, this kills me to hear it. RSS is the OG way to subscribe and share content out to others. It gives you control of what you subscribe to and reminds you when new content lands on your favorite sites. It is a huge component of web 1.0 and I feel is the most decentralized social media can ever hope to be.
I fully believe in our right to repair, ewaste reduction, and bringing a second life to still good hardware that is not up for it’s originally intended purpose. This is a sick console like experience you can strap to the back of a tv, throw in your back to take on a trip, or leave stuffed in your vehicle to game in the backseat. Sucks that it cant do 4k, but I’ve used mine on large screens, and it does quite well for a lot of games, maybe not AAA, but the cartoony multplayer games I play with my kids do quite well.
Damn these deepseek memes go hard. Wild to see openai get played by their own game.
It’s crazy that the normie news that I have seen on deepseek shows that the Chinese made what the Americans did at a fraction of the price, without taking notice that they are building on the shoulders of openai.
👏👏👏 This one is really good. I’m right there with him on most of this. I am very hesitant on subscription models, and all the ai tools feel like they are getting ready to be the next round of death by a thousand cuts, this time with pretty limited free tier and relatively high prices to run. I’m sure we will see companies get taken by huge bills soon by building off of someone else’s service.
On the flip side I’m definitely the guy that gets in a rut of just copy paste to the ai, wait for codeium to to inject. I feel like I have issues of momentum more than anything. When I’m on one side or the other I tend to stick it out for too long, but less so on going without because that llm drug is calling you when you hit a hard problem.
I’m excited to see him build out a homelab for llm stuff that he mentioned at the top. I’m interested, but probably not building one out for myself until we start to see some cheaper maybe used hardware to do it.
Kelsey says several times in this interview, you don’t need kubernetes. If you are running one node you don’t need kubernetes. My question though is, would you use kubernetes? Ya I get it if you are a web developer, data scientist, backend dev, but if you are looking to bee a whole ass engineer, or infrastructure engineer, you know kubernetes, Should you use kubernetes on single node?
I came accross from_attributes today it allows creation of pydantic models from objects such as a sqlalchemy Base Model or while nesting pydantic models. I believe in the past I have ran into some inconsistencies with nesting pydantic models and I’ll bet one had from_attributes set and another did not.
Arbitrary class instances¶ (Formerly known as “ORM Mode”/from_orm).
Pydantic models can also be created from arbitrary class instances by reading the instance > attributes corresponding to the model field names. One common application of this functionality is integration with object-relational mappings (ORMs).
To do this, set the from_attributes config value to True (see the documentation on Configuration for more details).
...
Dang strong stance that tmux and zellij should not exist. I really do get his point though. Theres a good number of terminal features I often miss out on because I run tmux. Its an app that runs apps, and doesn’t let all of the signals back to the host. But its fantastic at what it does, and brings so much to the table that the little bit of downside it brings is well worth it to me. The other thing missing in this discussion is that I can take my hotkeys and session workflow to any machine just by running tmux. I do not need to run a certain terminal, or install it headlessly on a server to get special features just for it.
Really good listicle of new modern top python libraries from 2024. Very well done article with images, links, and an actually quality listicle with many things I’ve never even heard of.
Good overview of seaborn color palettes. They have all sorts of different types, some designed to purposfully give each color the same weight for catecorization. Some designd to give linear differences in value, some have a parabolic feel with a diverging nature.
I’ve never seen or needed to use a serversideapply in kubernetes before, but I ran into this same issue in my k3s homelab while installing cloudnative-pg.
You can do it with argo
apiVersion: argoproj.io/v1alpha1 kind: Application spec: syncPolicy: syncOptions: - ServerSideApply=true
and you can do it with kubectl
I built out a tool for myself to manage my nvim configuration, and I wanted to quickly see which one I am running in my starship prompt. Here’s the config I ended up with. It warns if the NVIM_APPNAME environment variable is not set, and it shows which nvim I am using if it is set.