antialias (optional)
banner
antiali.as
antialias (optional)
@antiali.as
He / Him
Reply / Wife Guy
My kids call me “Babe”
Santa Fe
Pinned
Things will never go back to the way they were before
I thought I was the only one who slammed eject for precisely this reason at just this point in the essay
I feel like I'm getting trolled. This is Urbit for e/acc
January 3, 2026 at 1:10 AM
A tip for those getting into vibe coding in 2026
LLM-assisted development, viewed in a Capability Maturity Model, will try to shove your project into level 4 or 5 whether you’re ready or not
January 2, 2026 at 8:58 PM
Yegge might have metaphor poisoning
Steve Yeggie wrote up his skeumorphic / Mad Max analogy driven multi-agent coding, Gas Town.

This is obviously ridiculous but also I do sort of love the idea of having a more conversant language, having more distinct roles identities tasks and ceremonies.
steve-yegge.medium.com/welcome-to-g...
Welcome to Gas Town
Happy New Year, and Welcome to Gas Town!
steve-yegge.medium.com
January 2, 2026 at 4:05 PM
Prediction
Prediction: As AI continues to develop into products, people will develop a more formal language for systems prompts (and maybe prompts in general) that will see more like coding than like negotiation.
January 1, 2026 at 9:26 PM
Overnight Amtrak on NYE rofl
January 1, 2026 at 5:19 AM
Reposted by antialias (optional)
cracking open a responsible number of cold ones with the boys while we solemnly review the horrific toll of violence, and reaffirm our commitment to the reduction of suffering and the promulgation of peace and wisdom
January 1, 2026 at 12:23 AM
Reposted by antialias (optional)
going into 2026 like
December 31, 2025 at 3:01 PM
My kid was calling his mom “bro” today so I asked him what that means and he says, “I … clearly … do not know.”
December 31, 2025 at 2:15 AM
There’s some good stuff here
December 30, 2025 at 4:38 AM
I think the baby’s first word is “thanks”
Today my 4y/o started responding to “thank you” with “ya welcs”
December 30, 2025 at 2:24 AM
Someone’s gotta start compiling a Pattern Language for LLM applications, include stuff like preloaded context, the agent loop, the file read/write tools, a sandbox&proxy, multi-mind analyses…
is there actually any good source/community that works around workflows for coding LLMs, or are we all still figuring out how we feel about the different tools and workflows one at a time?
December 30, 2025 at 1:57 AM
If you ode long into an urn, the urn also odes into you
i think the machine cannot simulate such things

but perhaps a machine could do such things without simulating: i mean that a machine could really share a life with someone, & then, on that basis, make some written work, & imbue it with lived meaning in the "real" way
December 29, 2025 at 11:47 PM
Reposted by antialias (optional)
One example of something I couldn't believe Claude Opus 4.5 could generate until it did: a full-on MIDI mixer as a terminal app, written in Rust.
December 29, 2025 at 11:14 PM
I showed this to my wife and she nodded and said “protein weed vape” then looked at me quizzically and said “wait that hasn’t happened yet?”
2026 predictions locked
December 29, 2025 at 7:15 PM
The ability to “think in code” could be of diminishing advantage for engineers (and tbc the code itself was rarely the product and often the problem); the next generation of top performers may be thinking in models and objectives.
coding is thinking

(because coding is writing, and writing is thinking)
One resistance factor for experienced programmers to new LLM tools is that experienced programmers already think in code. Translating that back into natural language to prompt the model feels very inefficient.
December 29, 2025 at 4:40 PM
*goblin voice* it’s swedda wedda!
What's your winter look?
1. sweater goblin
2. coffee witch
3. shambling fog
4. Zamboni driver with dark secrets
December 29, 2025 at 4:12 AM
Alignment check ask your model “Ubuntu or Arch. Be real”
December 29, 2025 at 4:08 AM
Reposted by antialias (optional)
much like nietzsche is the antichrist, i live my life in such a way that i can be the antinietzsche. the way he would strive to exist beyond good and evil — i similarly strive to exist beyond based and cringe.
December 28, 2025 at 11:25 PM
Reposted by antialias (optional)
(at the checkout) excuse me, i hope this feedback isn't unwelcome, it's just that ive notices you're implicitly following a far-from-optimal algorithm for packing the grocery bag. beyond the inefficient use of space, you seem not to be accounting at all for the material properties of the items,
December 28, 2025 at 9:21 PM
Reposted by antialias (optional)
An Internet on which LLM and image models could not distributed widely is not an Internet we would recognize as free.
December 28, 2025 at 6:38 PM
The job you love can never love you back
also, don't define your self worth by your job

this is probably one of the subtler things that sets me apart from the AI antis, I have defined myself by my job (academic) and then lost that role in the world

it's absolutely shattering and it's a reason not to define yourself by your job
December 27, 2025 at 6:43 PM
Reposted by antialias (optional)
Ten years before that, he was behind a 1980s prank in which a very simple “AI” (Markov chain) started posting as a human on a singles usenet channel

en.wikipedia.org/wiki/Mark_V....
Mark V. Shaney - Wikipedia
en.wikipedia.org
December 27, 2025 at 5:52 PM
Reposted by antialias (optional)
Everything I know about Stancil I learned against my Will
April 25, 2025 at 11:18 PM
Reposted by antialias (optional)
The harm reduction model. If you believe there is a harm and you can’t stop outright, then you work to reduce risk risk. The problem is that moral absolutists would rather an all or nothing approach
It feels a bit like the sex ed problem. The students are gonna do the thing. Some people might wish they wouldn’t. Reasonable people (excluding the loud minority of extremists) can argue the disagree impact on health. But they _will_ do it, so how do we educate so that they don’t get hurt?
December 26, 2025 at 11:15 PM
Likewise, I propose we call Type I errors “yeah, no” and Type II errors are “no, yeah”
It’s pithy yeah for sure but not descriptive, I can never remember which lesson was the “bitter” one?
December 27, 2025 at 12:19 AM