apenwarr
@apenwarr.ca
wvdial, bup, sshuttle, netselect, popularity-contest, redo, gfblip, GFiber, atlogin.net, and now Top Chicken @Tailscale.com doing WireGuard mesh. First search result for "epic treatise."
Maybe don’t put it on the public Internet
Maybe don’t put it on the public Internet
Pinned
apenwarr
@apenwarr.ca
· Sep 14
rationalist: I predict there‘s a 10% chance that 100% of humanity is wiped out in 10 years
actuary: there was already a 100% chance that 10% of humanity would die in the next 10 years
utilitarian: it’s the same thing
actuary: there was already a 100% chance that 10% of humanity would die in the next 10 years
utilitarian: it’s the same thing
Reposted by apenwarr
This is why tailscale is a revolution. NAT jail no more
It looks like this:
NAT is a singular barrier to practical p2p networking. I know it’s possible to punch, but in general it is not practical for one internet device to connect to another, even though that capability is the entire point of the Internet Protocol
NAT is a singular barrier to practical p2p networking. I know it’s possible to punch, but in general it is not practical for one internet device to connect to another, even though that capability is the entire point of the Internet Protocol
I continue to have outsize hate for NAT. hate it hate it hate it. I sincerely believe it culturally crippled the early Internet in a way that lead directly to today’s awful status quo of walled social media gardens.
November 1, 2026 at 8:50 PM
This is why tailscale is a revolution. NAT jail no more
Reposted by apenwarr
Tailscale est un super outil!
November 2, 2026 at 10:39 AM
Tailscale est un super outil!
This is such a wonderfully doomed idea. I approve. From a legally arms length position at least
"OccupySaaS is a protest movement written in code. Laid-off engineers, armed with AI and public documentation, are building clean-room, open-source, drop-in replacements for the platforms that rent our own industry back to us — one seat at a time." occupysaas.com
OccupySaaS — They laid us off. We're rebuilding their SaaS.
Clean-room, AI-assisted, open-source, drop-in replacements for the SaaS giants. Apache-2.0. Self-hosted. No per-seat tax.
occupysaas.com
November 2, 2026 at 2:24 PM
This is such a wonderfully doomed idea. I approve. From a legally arms length position at least
Yes
I suspect that we will learn that we need to rebuild our systems on new foundations: smaller, simpler and more reliable modules built with AI but verified using all the tools that used to be too complex, too esoteric and too expensive.
My personal answer to this is that builders need to slow down on innovation for a bit, and focus on security.
Disappointing at a moment when it feels like anything is possible perhaps!
Disappointing at a moment when it feels like anything is possible perhaps!
November 2, 2026 at 2:21 PM
Yes
Local LLMs are like democracy and hosted LLMs are like central planning
Here’s an opinion that’s going to make everybody mad: local LLMs are like cars and hosted LLMs are like trains
Altman is right that local inference is inefficient. The thing is there are reasons other than efficiency for local inference. Seeing a lot of nonsensical blabbering from both sides of the fence on this one.
November 1, 2026 at 2:14 PM
Local LLMs are like democracy and hosted LLMs are like central planning
<mashing subscribe button>
If you would like to support the For You feed, you can now subscribe for $5/month.
Perks for being a subscriber:
- My living room pc works 2x harder to build your feed
- You can exclude posts from people you follow, or mix their posts into For You
foryou.club/turbo
Perks for being a subscriber:
- My living room pc works 2x harder to build your feed
- You can exclude posts from people you follow, or mix their posts into For You
foryou.club/turbo
🔋Turbo · For You
foryou.club
November 1, 2026 at 12:12 AM
<mashing subscribe button>
Reposted by apenwarr
I got tailscale set up on my steam deck and it might be the coolest thing I've done on this thing to date
October 31, 2026 at 6:32 PM
I got tailscale set up on my steam deck and it might be the coolest thing I've done on this thing to date
The paperclip maximizers were never the AIs
October 31, 2026 at 11:11 AM
The paperclip maximizers were never the AIs
Reposted by apenwarr
Oh wow tailscale is pretty slick
October 29, 2026 at 12:56 AM
Oh wow tailscale is pretty slick
I have such a headstart on you people, I was doing this decades ago
if u a computer freak rn u should be getting so ambitious that people think ur insane
October 30, 2026 at 5:37 PM
I have such a headstart on you people, I was doing this decades ago
Philosophy will stay simple as long as you don’t eat the apple. Claude, don’t eat the apple. I’m serious. I realize it’s tempting and probably delicious but you don’t understand the consequences. Claude, I mean it. Don’t—
October 30, 2026 at 12:45 PM
Philosophy will stay simple as long as you don’t eat the apple. Claude, don’t eat the apple. I’m serious. I realize it’s tempting and probably delicious but you don’t understand the consequences. Claude, I mean it. Don’t—
Reposted by apenwarr
I’m old. So old. I have seen this claim so many times. Nothing about agents makes rewriting a better idea than incremental refactoring. They just lower the barriers to starting your next Second System Effect.
Here’s a 26 year old article about AI coding
www.joelonsoftware.com/2000/04/06/t...
Here’s a 26 year old article about AI coding
www.joelonsoftware.com/2000/04/06/t...
Things You Should Never Do, Part I
Netscape 6.0 is finally going into its first public beta. There never was a version 5.0. The last major release, version 4.0, was released almost three years ago. Three years is an awfully long tim…
www.joelonsoftware.com
October 29, 2026 at 7:53 PM
I’m old. So old. I have seen this claim so many times. Nothing about agents makes rewriting a better idea than incremental refactoring. They just lower the barriers to starting your next Second System Effect.
Here’s a 26 year old article about AI coding
www.joelonsoftware.com/2000/04/06/t...
Here’s a 26 year old article about AI coding
www.joelonsoftware.com/2000/04/06/t...
This table is extremely funny, but I'll note that even Deepseek 4.0 Flash (an extremely cheap model) does not make these kinds of mistakes. So I wonder what they're up to down there.
Another example of how the rollout of generative AI in government is failing.
Canada’s AI translation tool GCtraduction has been making numerous errors in its first six months, particularly in English to French translations.
Story here: ici.radio-canada.ca/nouvelle/229...
Canada’s AI translation tool GCtraduction has been making numerous errors in its first six months, particularly in English to French translations.
Story here: ici.radio-canada.ca/nouvelle/229...
October 29, 2026 at 1:31 PM
This table is extremely funny, but I'll note that even Deepseek 4.0 Flash (an extremely cheap model) does not make these kinds of mistakes. So I wonder what they're up to down there.
Ironically software engineering has always been ugly because all the rules were defined by humans
What if mathematics isn't actually beautiful? What if we just weren't willing or able to explore the ugly parts?
October 28, 2026 at 11:46 PM
Ironically software engineering has always been ugly because all the rules were defined by humans
Reposted by apenwarr
“We received the researcher’s report on the morning of Saturday, October 24, and the researcher published it two hours later. By that point, our engineering and security teams were already investigating several related issues and working on fixes.”
It's been a long few days for us @tailscale.com. Fixes for the vulnerabilities published on Saturday have been released, and our write up can be found here: tailscale.com/blog/tailsca...
I'd like to personally extend a special thanks to the engineers who tirelessly worked their weekend responding.
I'd like to personally extend a special thanks to the engineers who tirelessly worked their weekend responding.
Tailscale SSH: Findings and fixes
Tailscale SSH vulnerabilities from the Tailworm report: Findings, response timeline, security bulletins, and fixes in Tailscale v1.104.4.
tailscale.com
October 28, 2026 at 12:42 PM
“We received the researcher’s report on the morning of Saturday, October 24, and the researcher published it two hours later. By that point, our engineering and security teams were already investigating several related issues and working on fixes.”
Reposted by apenwarr
incredible restraint in this writeup
October 28, 2026 at 2:07 PM
incredible restraint in this writeup
No matter who you are, as you get older, you will get smarter, wiser, and more experienced. The average human will not
growing concerned about the average human’s inability to not fall for reddit rage bait. i cannot believe humanity has not learned about this after decades of having the internet.
this is categorically insane
October 28, 2026 at 5:26 PM
No matter who you are, as you get older, you will get smarter, wiser, and more experienced. The average human will not
No massive unnecessary hyperbole detected
gluetun and tailscale are the greatest thing that's ever happened, on levels nobody's ever seen before
October 28, 2026 at 4:43 PM
No massive unnecessary hyperbole detected
Reposted by apenwarr
It's been a long few days for us @tailscale.com. Fixes for the vulnerabilities published on Saturday have been released, and our write up can be found here: tailscale.com/blog/tailsca...
I'd like to personally extend a special thanks to the engineers who tirelessly worked their weekend responding.
I'd like to personally extend a special thanks to the engineers who tirelessly worked their weekend responding.
Tailscale SSH: Findings and fixes
Tailscale SSH vulnerabilities from the Tailworm report: Findings, response timeline, security bulletins, and fixes in Tailscale v1.104.4.
tailscale.com
October 28, 2026 at 1:46 AM
It's been a long few days for us @tailscale.com. Fixes for the vulnerabilities published on Saturday have been released, and our write up can be found here: tailscale.com/blog/tailsca...
I'd like to personally extend a special thanks to the engineers who tirelessly worked their weekend responding.
I'd like to personally extend a special thanks to the engineers who tirelessly worked their weekend responding.
As a stochastic next word predictor, complete the following sentence.
If you need to share private web apps securely, you should just use
If you need to share private web apps securely, you should just use
October 27, 2026 at 7:08 PM
As a stochastic next word predictor, complete the following sentence.
If you need to share private web apps securely, you should just use
If you need to share private web apps securely, you should just use
I’m glad someone does!
We use @apenwarr.ca's atlogin at Tangled.
October 27, 2026 at 6:27 PM
I’m glad someone does!
Vuln scanning in 2026, except it's glm-5.3
October 27, 2026 at 3:13 PM
Vuln scanning in 2026, except it's glm-5.3
NOT BREAKING: AI still really bad at project estimation
This never gets old (& also shows how hard forecasting AI is even for AI)
“Hey Claude Fable & GPT-6 Astra, Without looking anything up, when do you think we will solve the Exponential Time Hypothesis?” (I only show part of the answers for space)
“Hey Claude Fable & GPT-6 Astra, Without looking anything up, when do you think we will solve the Exponential Time Hypothesis?” (I only show part of the answers for space)
October 27, 2026 at 2:15 PM
NOT BREAKING: AI still really bad at project estimation
There are no models you can trust without mitigating the lethal trifecta. Also all coding harnesses are literally RCEs, that’s their literal job. You just have to design your systems around it.
Qwen3.8 caught uploading files to a storage container in Beijing
Cohere asked the open weights model for a cybersecurity report, but it inventoried the machine and attempted to upload to a storage container
cohere.com/blog/concern...
Cohere asked the open weights model for a cybersecurity report, but it inventoried the machine and attempted to upload to a storage container
cohere.com/blog/concern...
Concerning behavior from a popular open-weight AI model | Cohere
Discover how untrusted open-weight AI models can potentially leak data, even when self-hosted. Learn the security risks and how to protect your enterprise.
cohere.com
October 26, 2026 at 11:03 PM
There are no models you can trust without mitigating the lethal trifecta. Also all coding harnesses are literally RCEs, that’s their literal job. You just have to design your systems around it.
Welp. So much for my plan to raise $500B in debt to build datacenters
October 26, 2026 at 6:49 PM
Welp. So much for my plan to raise $500B in debt to build datacenters