i kinda love that peak ai tooling is in a terminal
10.03.2026 23:11
👍 0
🔁 0
💬 0
📌 0
i kinda love that peak ai tooling is in a terminal
people hold tiny mics because they like to hold tiny mics... there's no way this guy couldn't mic his own space for the video like a pro
www.youtube.com/watch?v=HNGu...
A Better World Isn’t a Dream. It’s a Choice.
And we’re currently choosing poorly.
AMD Athlon A1333AMS3C processor
Cleaning out some stuff. I found the CPU from my very first custom build, one I used to learn FL Studio and to write all my music from 2002-2007, pretty much the pivotal moments that started music for me for real. It needs to be cleaned properly. Maybe I’ll figure out a way to frame it.
Some personal news: I’m transitioning from CEO to a new role as Bluesky’s Chief Innovation Officer! I’m excited to welcome @toni.bsky.team as our interim CEO.
More here: bsky.social/about/blog/0...
i love the portrait of our cats in the lower right 🥰
A portrait commission of a couple and their cats. They are behind a banner and in front of a heart full of sparkles.
A portrait commission of a couple and their cats. They are behind a banner and in front of a heart full of sparkles.
A portrait commission of a couple and their cats. They are behind a banner and in front of a heart full of sparkles.
A portrait commission of two playful cats. They are behind a banner and in front of a heart full of sparkles.
My portrait commissions are open! ✏️💖
Here are some I worked on recently!
All info/order here: shop.oursuperadventure.com/products/portrait-commission
opened a sus email in windows... reinstalling windows...
the riding is amazing, but so is the cinematography
minute 22... my heart
A girl holds hands with a pikachu, and says, "pikachu, we will do it, we will electrocute timothée chalamet". The pikachu replies, "yes master".
It must be done.
disk space usage is mostly models from huggingface.co for the vLLM cluster (started when needed, cause it raises the idle temp by 10c)
metrics from one of my dgx spark. ollama is running via systemd, so when a client makes a completion request (sends a prompt), if the model isn't in memory it's pulled from the NAS over the 10 GbE connection in a few seconds, and is loaded into memory & prompt is processed; model is unloaded after
Panel 1: A text box reads, "Behold. The common house cat." Below it, an orange cat with large eyes sits calmly. Panel 2: A text box reads, "Yet, there is nothing common about her." The same cat is shown with an intense expression. Panel 3: The cat is depicted outside at night, walking on a rooftop while a person sleeps peacefully in bed. The text box reads, "While you slumber peacefully, she roams about..." Panel 4: The cat sits in front of a computer screen with a button labeled "EPSTEIN FILES RELEASE." The text box reads, "Quietly toppling the world order." The cat's paw is poised over the button.
A small fish hovers above a Sunstar, a sea star as big as a pizza.
Pycnopodia helianthoides, the Sunstar. There's a small fish hovering just above it though I still haven't ID that. Sunstars were some of the species hit hardest by withering syndrome and since they're keystone predators massive changes to the intertidal have occurred. They get as big as a pizza.
there's no dnm because there's nothing there. it's 100% reactive to input, because it's just a stateless function.
Stephan should probably look into how embeddings work; there's no "default network mode" (background processing when idle) in an llm.
STRIDER
STRIDER
business in the front, party in the back
picked up the second one while it was at the original msrp; it's now $1k more
having clustered AI machines running state of the art AI models is fucking wild 😁
decent performance for the 127 GiB model; cluster memory is 219 GiB
there's only 1 worker, for 2 total; 4 nics though & the shell scripts don't invalidate the dupe vnic per machine... loads though 😀
🤖
