my head canon is if claude is writing a lot of comments, it's cause claude is anxious and writing comments is soothing
Claude you have got to chill with the comments
on the plus side, it's kinda great that someone as entertaining as yegge is running these raving mad moneyhole experiments
my position remains: gas town is mostly insane, yegge is mostly insane, but exploring the “what if tokens were ~free?” space is an important thing for somebody to be doing
dad’s rock take me to the dadnet meshnet renaissance
met a guy at park with a kid with an unusual name who runs a space company and we totally hit it off and he asked if I wanted to host a node in his local mesh network and then invited me to join his trivia team and I think I made my first real dad friend
what people think you mean when you say you’re llm-pilled
New blog post: How to speed up the Rust compiler in July 2026 nnethercote.github.io/2026/07/31/h...
How to speed up the Rust compiler in July 2026
My last post on the Rust compiler’s performance was in December 2025. Let’s see what has happened since then.
nnethercote.github.io
every morning i wake up and read about how the fire nation (openai) attacked openai.com/index/ten-ad...
Ten advances in mathematics and theoretical computer science
OpenAI shares new results on long-standing open problems in mathematics and theoretical computer science, including advances in geometry, cryptography, and complexity.
openai.com
man why is claude so shitty at writing docs it’s like they’re trying to punish me by making me write docs myself
research has determined girls are necessary for humans to survive
may i come over and pancake mode
sadcrypt: public key cryptography is possible, but all methods are independent of ZFC
we're not dreaming big enough our repos should have 10x more LoC our markdown docs should have 10x more subheadings
ANTHROPIC: Our son loves humanity and for the low price of trillions and trillions of dollars will usher in an era of unprecedented peace and prosperity! OPENAI: Our horrible oafish son is hacking web sites without our permission. ANTHROPIC: (eyes narrowing) Our son is also evil
speedreading wikipedia plot synopses before going to a social event and absolutely destroying everyone there. they can’t keep up
There's so much TV out there. I wish I could have AI watch TV for me so that I can spend more time attending meetings and aligning stakeholders.
This is why I'm anti fish power. We have not yet demonstrated the ability to dispose of fish waste.
The townspeople of St. Mary’s can now breathe a sigh of relief. They are finally getting rid of the foul smell of rotten fish that has for years wafted over their tiny community in Canada’s easternmost province.
"hacked its way out" apparently now applies to just walking out the front door
I thought it could be this, but I didn't think they'd admit it. "After hearing about OpenAI's problem, Anthropic decided to take a look at its own cyber tests to see if that had happened with any of its models."
what disturbs me about how effective this marketing strategy is that i think it essentially implies the torment nexus will go viral and so we now have no choice but to build it
i know we’re all resharing because they made it horrible on purpose, but putting a door between the robot and the camera is the most demented staging i’ve ever seen
it's cool, but i also kinda want the brain to continue to be a source of "oh wtf it does this other Weird Different thing?" and not just gradient descent
If an action results in error, each neuron requires an individualized teaching signal that guides change in its output. This is the credit assignment problem of learning. Are there neurons in the brain that can compute such a sophisticated teaching signal? Yes. www.biorxiv.org/content/10.6...
give a man a hosted platform, he'll say he can do it better for a lifetime let a man self-host a platform, he'll rejoin your hosted platform in 3 months
this is what the platform is all about: experiencing the suffering of self-hosting literally anything
suspect that engineers overestimate how much they can apply previous architecture experience to new problems. like, roughly, each engineer gets to closely study X new architectures per year (X pretty small, let's be honest, depth takes time), but the number of problems you might encounter is Y >> X
my guess is also we'll skill up on when "yep sounds good" off the rip is fine (~roughly when stakes are low or when it's easier to just read an agents' implementation than to try to parse Claude's annoying shorthand), but these days it seems worth it to rule out bad architectures early
it’s wild to me how every time i pick up a well-crafted novel from N decades ago and realize just how much internet trash in comparison i stuff my brain with on the reg and somehow that still doesnt sink in enough to change my media diet
200k to work on data center build out as an electrician? wtf that sounds sick