apenwarr
banner
apenwarr.ca
apenwarr
@apenwarr.ca
wvdial, bup, sshuttle, netselect, popularity-contest, redo, gfblip, GFiber, and now CEO @Tailscale.com doing WireGuard mesh. Top search result for "epic treatise."
Pinned
So I’m being told that “we put the net in skynet” is not a good motto for some obscure reason
There are things that empirically work and things that empirically don’t work. I’m often surprised at how much trouble people have in telling the difference.
November 26, 2025 at 12:34 AM
Sorry, immune system, I'm afraid you're not allowed to get sick for another two weeks. I can slot you in for December 10th, does that work? Great, thanks.
November 25, 2025 at 3:18 PM
Do you think when Gemini uses Google search, it gets a bunch of AI results on top
November 25, 2025 at 12:22 PM
Discussed with Gemini how to create synthetic surround sound from stereo music. Very educational, I learned a lot about speaker placement etc.

But it completely just lied and fabricated some of the key steps to actually *doing* the signal processing. Just like asking on a forum! But faster.
November 24, 2025 at 3:33 PM
so much happens
November 23, 2025 at 11:48 PM
Waiting for the impossible to stop happening so that whatever remains, however improbable, can happen instead
November 23, 2025 at 2:52 AM
Reposted by apenwarr
If only the lovely @teiconsortium.bsky.social folks had used JSON instead of SGML + XML, dozens of knowledge workers might've been employable more widely! (Joke, mostly--I knew TEI P3-mid P5.)

@apenwarr.ca's connectivity tale is pretty solid re how + when things opened up
apenwarr.ca/log/20251120
Systems design 3: LLMs and the semantic revolution
Long ago in the 1990s when I was in high school, my chemistry+physics teacher pulled me aside. "Avery, you know how the Internet works, righ...
apenwarr.ca
November 21, 2025 at 11:11 PM
Reposted by apenwarr
Sounds great. How about we expand it to all AI companies as well? Or like ... maybe just every tech company that isn't Tailscale.
npr.org NPR @npr.org · 5d
Sens. Mark Kelly, D-Ariz.,and John Curtis, R-Utah, want to hold social media companies accountable for the negative impacts their algorithms have on people. They spoke to NPR about their bill. n.pr/3JZlY6o
2 Senators want to hold big tech accountable for harms caused by algorithms
Sens. Mark Kelly, D-Ariz.,and John Curtis, R-Utah, want to hold social media companies accountable for the negative impacts their algorithms have on people. They spoke to NPR about their bill.
n.pr
November 20, 2025 at 4:47 PM
Countries, as measured by their airport wifi

Lisbon: doesn’t work, 404 not found
Zürich: chargea money
Helsinki: just works, no nonsense
November 20, 2025 at 3:26 PM
A status update on my ongoing ChatGPT product naming chat:

"These feel slightly too epic or narrative-heavy, like installing a creation myth."
November 20, 2025 at 11:54 AM
Ironically the word enshittification has been enshittified by its own users, far faster than any of the products it was meant to explain.
November 20, 2025 at 10:32 AM
Someday I hope to achieve this level of optimistic outlook
Today's weather has a lot of positives actually
November 20, 2025 at 10:20 AM
What is “executive misalignment?”
November 20, 2025 at 6:58 AM
#til that not only do Atlantic salmon and Pacific salmon exist, but there is also North Sea salmon and it’s awesome. Thanks Finland. See you next time!
November 19, 2025 at 9:29 PM
And also, here’s your periodic reminder that Baby Shark is technically K-Pop
A friend of mine has early access to cutting edge corporate jargon, I heard the phrase "let's double-click on that" from him long before anywhere else. I asked him what's new these days and he says it's "the shark closest to your body" for the most urgent issue.
November 19, 2025 at 8:55 PM
Arriving in Finland. Seems misnamed. Lots of Finns, hardly any fins. Certainly not on land
November 18, 2025 at 6:02 PM
Idea: a decentralized data plane
November 18, 2025 at 5:54 PM
Reposted by apenwarr
the two types of organizations: spreadsheets and cults
November 17, 2025 at 5:52 AM
I like nuance and I like that this thread adds some.
My mistake has been trying to explain what "fascism" is to people who don't care what it is.

So I'm going to try something else. Let's do this as a practical thought exercise.

If America *were* fascist *today* - what would you have to do differently in your daily life? /1
November 16, 2025 at 11:57 PM
Reposted by apenwarr
dont sleep on fusion. fusion has the juice. look at this shit. impeccable sci fi vibe. alien curvature for reasons u wouldnt understand bc u dont know what poincare sections of the plasma beam are. it's even greebled with ports and domes and shit. u can tell the physics nerds are really cooking here
November 16, 2025 at 5:06 AM
As a chicken, I
November 15, 2025 at 12:53 PM
I predict that we‘ll invent ai telepathy, and the mechanics of it will be pretty straightforward (copy parts of compatible LLM state over Ethernet, bypassing the language bit). And I’m looking forward to this updating the debate about human telepathy.
November 15, 2025 at 10:46 AM
Hmm, 5.3x boost in 3-ish years. Not too far off Moore’s Law.

As I like to say, Moore’s Law is still with us, it just keeps getting weirder.
Research shows local language models handle 88.7% of queries, cutting dependence on cloud infrastructure. With a 5.3x boost in energy efficiency since 2023, local AI transforms large-scale processing and conserves resources. https://arxiv.org/abs/2511.07885
Intelligence per Watt: Measuring Intelligence Efficiency of Local AI
ArXiv link for Intelligence per Watt: Measuring Intelligence Efficiency of Local AI
arxiv.org
November 13, 2025 at 8:25 PM
Tailscale continies its trend of being perfect for AI despite being 0% optimized for AI
It turns out that Tailscale is really perfect for how I use the Spark.

Almost nothing I do is graphical, but I’ve already found myself running multiple little ML services on it that I can just… use from my phone, wherever.

As well as just being able to SSH into it from my laptop for dev purposes.
The tiny new computer is setting itself up
November 11, 2025 at 11:33 PM
Sooooo, do I know anybody in Lisbon this week?
November 10, 2025 at 11:28 AM