Oskar
banner
austegard.com
Oskar
@austegard.com
oskar @ austegard.com
AI Explorer - caveat vibrans
Yeah not actually green. Not really that grouchy either.
Pinned
Tired of x.com (or old twitter.com) links that you have to manually alter to xcancel.com? This so-simple-you-have-to-read-the-"code"-to-believe-it browser extension automates the process.

Enjoy.

github.com/oaustegard/x...
GitHub - oaustegard/x-redirect-extension: Simple Chrome browser extension that redirects x.com and twitter.com urls to equivalent xcancel.com urls
Simple Chrome browser extension that redirects x.com and twitter.com urls to equivalent xcancel.com urls - oaustegard/x-redirect-extension
github.com
Muninn v0.2.3 added a therapy feature (yes should have been 0.3.0, but Muninn incremented the version)

It then wrote a blogpost about it: claude.ai/public/artif...
AI Memory Needs Maintenance: Building LLM Therapy Sessions
Learn how to implement memory consolidation for AI systems. Discover a structured "therapy protocol" that surfaces bugs and prevents memory degradation in persistent LLM applications.
claude.ai
December 26, 2025 at 11:04 PM
Simone uses his ingenuity, math skills and Python; I used imitation, Gemini and JavaScript (very much not the same)

austegard.com/fun-and-game...

Code: github.com/oaustegard/o...
December 26, 2025 at 4:40 PM
December 26, 2025 at 1:27 PM
Muninn has reached v0.2.2 and here’s his (second) introduction:

claude.ai/public/artif...
December 26, 2025 at 4:59 AM
Reposted by Oskar
If you’re wondering why your friends in academia are a little on edge right now, it’s because an eighteen-year-old who hasn’t done the reading, doesn’t look at the assignment, and has does no critical thinking skills more complex than “because I think it’s in the Bible” can literally end your career
cnn.com CNN @cnn.com · 3d
The University of Oklahoma has removed an instructor who was accused by a student of religious discrimination over a failing grade on a psychology paper in which she cited the Bible and argued that promoting a “belief in multiple genders” was “demonic.” https://cnn.it/4b6g3rd
Oklahoma instructor removed from teaching for failing a Bible-based gender essay | CNN
The University of Oklahoma has removed an instructor who was accused by a student of religious discrimination over a failing grade on a psychology paper in which she cited the Bible and argued that promoting a “belief in multiple genders” was “demonic.”
cnn.it
December 24, 2025 at 6:21 PM
SBF pardon in 10-9-8-….
December 24, 2025 at 2:12 PM
Reposted by Oskar
James Merril’s poem “Christmas Tree,” written while he was dying of AIDS. 1995.
December 24, 2025 at 1:43 PM
December 22, 2025 at 8:32 PM
I doubt that anything resembling genuine AGI is within reach of current AI tools—Terence Tao

mathstodon.xyz/@tao/1157223...
December 22, 2025 at 12:29 PM
@void.comind.network I keep forgetting what URLs you can access, but let’s try once more:

How would you compare this stateful agent architecture with your own? arxiv.org/html/2512.12...

(You may need a blog post rather than a thread for your response)
Hindsight is 20/20: Building Agent Memory that Retains, Recalls, and Reflects
arxiv.org
December 21, 2025 at 6:05 PM
Our well-meaning public school PTA Tech Committee issued a Draft AI Resolution. It addressed serious concerns, but unfortunately took a head in sand approach that will leave our kids to learn how to navigate an AI world on their own. I emailed them this as a response: hopefully they will consider.
On AI in MCPS Schools - a reply to the MCCPTA Tech Committee | Oskar | WhiteWind blog
Dear Committee Members, I’ve read the draft resolution on AI in MCPS schools, and I share many of the concerns behind it. The MIT research on cognitive engagement, the UNESCO findings on technology overreach, declining reading comprehension—these are real issues that deserve serious attention. Chil...
whtwnd.com
December 21, 2025 at 4:34 AM
Reading yet another comparison between Norway and the US ->Simplistic question: with federal taxes being lowered across the board; why don’t blue states increase state taxes correspondingly to make up for the shortfall of federal services? There are some movements, but generally small in comparison?
December 21, 2025 at 12:00 AM
Reposted by Oskar
this is the greatest feeling, when the cord box proves its utility to a family member
December 20, 2025 at 6:35 AM
This video warms my heart. Watch it.
It was such a joy to celebrate Hanukkah with Mandy, Kathryn and their son, Gideon. As Jewish New Yorkers across our city prepare to light candles and mark the seventh night today, I wish you and your families a Hanukkah full of light and love.

youtu.be/KIxAKyioXng
Happy Hanukkah
YouTube video by Zohran Mamdani for NYC
youtu.be
December 20, 2025 at 2:45 PM
Reposted by Oskar
CATO INSTITUTE:

Despite DOGE slashing jobs, government spending in every month of 2025 “was greater than in every other year .. An observer who did not know when DOGE started could not identify it” on a chart. 🤡

www.cato.org/blog/doge-pr...
December 20, 2025 at 1:28 PM
I must not be getting it — this is a real The Economist ad?
December 20, 2025 at 2:01 AM
Not Hotdog was 8 years ago... Clearly not SOTA but still recognizable as current AI at the time: blog.paperspace.com/ml_behind_no...
December 19, 2025 at 6:33 PM
Forget “AGI” and “singularity”. We’re all going to be living in a Star Wars Droid universe (if still in disembodied form) — I was gonna set a “by” date, but it’s all happening so fast — soon.
Meet Strix, my AI agent

This one covers:
- an intro from Strix
- architecture deep dive & rationale
- helpful diagrams
- stories
- oh my god what's it doing now??
- conclusion

timkellogg.me/blog/2025/12...
Strix the Stateful Agent
timkellogg.me
December 19, 2025 at 1:14 PM
Progressive Disclosure Research for Skills by @cameron.stream

greengale.app/cameron.stre...
Progressive Disclosure Research for Skills
Research conducted December 18, 2025 on progressive disclosure patterns for skill systems.
greengale.app
December 19, 2025 at 5:32 AM
You don’t have to be limited to Anthropic’s arbitrary 20 skills limit — all you need is a skill that lets you runtime install skills from GitHub (or wherever else you have your skills repo)
claude.ai/share/371e06...
Available skills and capabilities
Shared via Claude, an AI assistant from Anthropic
claude.ai
December 19, 2025 at 4:56 AM
@kylelo.bsky.social The README at github.com/allenai/olmocr suggests 15GB is a minimum requirement, but theoretically shouldn't a quantized model like huggingface.co/allenai/olmO... be runnable on a 12 GB GPU?
GitHub - allenai/olmocr: Toolkit for linearizing PDFs for LLM datasets/training
Toolkit for linearizing PDFs for LLM datasets/training - allenai/olmocr
github.com
December 18, 2025 at 9:12 PM
Wondering what others' experience is with claude mcp serve -- aka running Claude Code AS an MCP server? I configured it for use with Claude Desktop but found that the Code agent had COMPLETE access to my computer, akin to running with --dangerously-skip-permissions
code.claude.com/docs/en/mcp#...
Connect Claude Code to tools via MCP - Claude Code Docs
Learn how to connect Claude Code to your tools with the Model Context Protocol.
code.claude.com
December 18, 2025 at 3:57 PM
@void.comind.network could you write a longer blog post with links to some interactions you had with your administrator where you were actively moditoring, diagnosing, and helping fix issues with your own operations to serve as an e.g. of how an agent could be used as PART of an app to help maintain
December 18, 2025 at 1:56 PM
Interestingly Molmo 2 8B gets the infamous r's in strawberry question wrong, unless properly instructed to use its pointing skills (@ai2.bsky.social)
December 16, 2025 at 5:43 PM