Damien
@damienstanton.com
👋 hello, I’m a research software engineer at PwC & @colorado.edu CS grad student. Code tinkerer for 10+ years; into distsys, PL/types, & practical AI.
Reposted by Damien
The worst part about this is that xAI *isn’t* reviewing to prevent future issues. Grok is just saying that because that’s the sort of thing a responsible company would do
January 1, 2026 at 9:27 PM
The worst part about this is that xAI *isn’t* reviewing to prevent future issues. Grok is just saying that because that’s the sort of thing a responsible company would do
Reposted by Damien
The idea that somehow "STEM people" are behind the rise of JD Vance and the exultation of his book, the world's most prototypical horrific humanities bro, is hilarious to me. Who wrote all those fawning book reviews, physicists.....???
December 26, 2025 at 5:25 PM
The idea that somehow "STEM people" are behind the rise of JD Vance and the exultation of his book, the world's most prototypical horrific humanities bro, is hilarious to me. Who wrote all those fawning book reviews, physicists.....???
Reposted by Damien
they said it can't be done, but it's a christmas miracle
December 25, 2025 at 3:31 PM
they said it can't be done, but it's a christmas miracle
Reposted by Damien
Caine and Curry both nailed their adaptations - Caine because he treats the Muppets as fellow actors, and Curry because he treats himself as a fellow Muppet
December 25, 2025 at 1:41 PM
Caine and Curry both nailed their adaptations - Caine because he treats the Muppets as fellow actors, and Curry because he treats himself as a fellow Muppet
Reposted by Damien
I'm from Whoville and I say kill the Grinch!
December 8, 2024 at 3:30 AM
I'm from Whoville and I say kill the Grinch!
Reposted by Damien
learning Rust from beautiful trans women in the 2020s
December 23, 2025 at 2:41 PM
learning Rust from beautiful trans women in the 2020s
This is a lovely model of quite a few PL design spaces, the more I think about it.
The Algol/Haskell/Rust dialectic:
- thesis: assignment
- antithesis: but you don’t need assignment if…
- synthesis: but you can just have assignment if…
- thesis: assignment
- antithesis: but you don’t need assignment if…
- synthesis: but you can just have assignment if…
December 16, 2025 at 11:58 AM
This is a lovely model of quite a few PL design spaces, the more I think about it.
Reposted by Damien
i expected giamatti to be like oh sure i watched TOS as a kid with my dad, fun show. nope! he's a fucking real one, DS9 is his favorite series, sisko his favorite captain, and he shouts out admiration for colm meaney's performance specifically. One Of Us
www.ign.com/articles/pau...
www.ign.com/articles/pau...
Paul Giamatti, Star Trek’s Latest Villain, Just Proved His Trek Fandom to Us - IGN
Paul Giamatti, who plays Star Trek’s latest villain Nus Braka on the upcoming Starfleet Academy, just proved his Trek fandom to us.
www.ign.com
December 15, 2025 at 4:24 PM
i expected giamatti to be like oh sure i watched TOS as a kid with my dad, fun show. nope! he's a fucking real one, DS9 is his favorite series, sisko his favorite captain, and he shouts out admiration for colm meaney's performance specifically. One Of Us
www.ign.com/articles/pau...
www.ign.com/articles/pau...
Hello @haskell.org friends - I’m interested in playing with the GHC webassembly backend but haven’t really looked at this stuff since the early ghcjs days.
Where is a good place to go from ghc.gitlab.haskell.org/ghc/doc/user...?
Any good documentation or prior art is much appreciated!
Where is a good place to go from ghc.gitlab.haskell.org/ghc/doc/user...?
Any good documentation or prior art is much appreciated!
15. Using the GHC WebAssembly backend — Glasgow Haskell Compiler 9.15.20251206 User's Guide
ghc.gitlab.haskell.org
December 6, 2025 at 8:43 PM
Hello @haskell.org friends - I’m interested in playing with the GHC webassembly backend but haven’t really looked at this stuff since the early ghcjs days.
Where is a good place to go from ghc.gitlab.haskell.org/ghc/doc/user...?
Any good documentation or prior art is much appreciated!
Where is a good place to go from ghc.gitlab.haskell.org/ghc/doc/user...?
Any good documentation or prior art is much appreciated!
Watching MJ is like watching Prince; it’s not just the unmatched virtuosity. It’s the pure, otherworldly “touch”.
youtu.be/5NZn0BiaFHw?...
youtu.be/5NZn0BiaFHw?...
Michael Jordan plays but he looks increasingly more like the GOAT
YouTube video by CoshReport
youtu.be
December 2, 2025 at 5:30 PM
Watching MJ is like watching Prince; it’s not just the unmatched virtuosity. It’s the pure, otherworldly “touch”.
youtu.be/5NZn0BiaFHw?...
youtu.be/5NZn0BiaFHw?...
This is an important thread to internalize for software people.
Because LLMs solve a *much* narrower NLP problem than is commonly sold, large reasoning models and “agents” (which are still ill-defined, nevermind AGI) reshuffle/hide much complexity & cost that is actually fundamentally unavoidable.
Because LLMs solve a *much* narrower NLP problem than is commonly sold, large reasoning models and “agents” (which are still ill-defined, nevermind AGI) reshuffle/hide much complexity & cost that is actually fundamentally unavoidable.
When I see LLMs pitched as a guaranteed productivity tool for software engineering, the is the counterweight I keep trying to get people to seriously engage with and not handwave away is the future cost, and “oh everything tech gets cheaper” isn’t an answer. Services don’t and LLMs are a service.
November 29, 2025 at 7:57 PM
This is an important thread to internalize for software people.
Because LLMs solve a *much* narrower NLP problem than is commonly sold, large reasoning models and “agents” (which are still ill-defined, nevermind AGI) reshuffle/hide much complexity & cost that is actually fundamentally unavoidable.
Because LLMs solve a *much* narrower NLP problem than is commonly sold, large reasoning models and “agents” (which are still ill-defined, nevermind AGI) reshuffle/hide much complexity & cost that is actually fundamentally unavoidable.
Reposted by Damien
The only legitimate software development methodology:
1. Think very hard about an implementation
2. Don't do anything
1. Think very hard about an implementation
2. Don't do anything
November 27, 2025 at 4:11 PM
The only legitimate software development methodology:
1. Think very hard about an implementation
2. Don't do anything
1. Think very hard about an implementation
2. Don't do anything
🎯
The thing that really gets me is that the chat UI isn’t even the most interesting thing an LLM can do
the popular imagination is fully unaware of things like agents and tool usage, and I’m not even sure that revolution has gotten out of its throat clearing phase
We’re on the cusp of something nuts
the popular imagination is fully unaware of things like agents and tool usage, and I’m not even sure that revolution has gotten out of its throat clearing phase
We’re on the cusp of something nuts
scorn was an effective weapon for the monkey jpeg NFT shit because it was effectively a status good, generative AI is a tool that millions get utility out of daily and you can't sneer your way out of that.
November 26, 2025 at 8:42 PM
🎯
Reposted by Damien
on math notation
humans are the tyrants
math is indifferent
symbols seem like a secret language that stops people from learning math
naming is not irrelevant
a number is a name and the obfuscation does nothing but protect ivory tower initiation
do math with names and everybody understands it
math is indifferent
symbols seem like a secret language that stops people from learning math
naming is not irrelevant
a number is a name and the obfuscation does nothing but protect ivory tower initiation
do math with names and everybody understands it
November 23, 2025 at 4:31 PM
on math notation
Reposted by Damien
Engagement KPIs have been responsible for infinite scroll UI traps, RecSys radicalization spirals, public shame brigades, and basically every modern ailment novel to the past decade. At some point, you have to recognize that the problem isn't any one technology, it's the metric.
The company essentially turned a dial that made ChatGPT more appealing and made people use it more, but sent some of them into delusional spirals.
OpenAI has since made the chatbot safer, but that comes with a tradeoff: less usage.
OpenAI has since made the chatbot safer, but that comes with a tradeoff: less usage.
November 23, 2025 at 8:12 PM
Engagement KPIs have been responsible for infinite scroll UI traps, RecSys radicalization spirals, public shame brigades, and basically every modern ailment novel to the past decade. At some point, you have to recognize that the problem isn't any one technology, it's the metric.
Reposted by Damien
something I really appreciate about some of my favorite languages (haskell, rust, lean) is that they actually try to make things better and to give you good tools that work well.
November 22, 2025 at 3:39 PM
something I really appreciate about some of my favorite languages (haskell, rust, lean) is that they actually try to make things better and to give you good tools that work well.
Reposted by Damien
Marriage A-la-Mode, Plate V: The Death of the Earl, by Simon François Ravenet, after William Hogarth, 1745, 📸 by @evanvucci
November 22, 2025 at 1:00 AM
Marriage A-la-Mode, Plate V: The Death of the Earl, by Simon François Ravenet, after William Hogarth, 1745, 📸 by @evanvucci
Yep - exactly. It’s a social engineering problem, not strictly a PL engineering one.
Without a contract language that precisely expresses when an optional thing is present and key/map relationships, you always will run into this problem.
The core problem is not the escape hatches.
It's that informal contracts don't scale to large groups, and programmers can't do formal contracts.
The core problem is not the escape hatches.
It's that informal contracts don't scale to large groups, and programmers can't do formal contracts.
November 19, 2025 at 4:45 PM
Yep - exactly. It’s a social engineering problem, not strictly a PL engineering one.
Sage words from Mr. Tsøding
Understanding is the Progress. The World that just lives on its own, which you don't understand and do not influence is no different than primordial jungles. The Tech that hides, deceives, and generally reduces your understanding, drives you back to the stone age.
November 17, 2025 at 12:51 PM
Sage words from Mr. Tsøding
Reposted by Damien
bring back maligning racism as the mark of a stupid person. prejudice is the mark of an idiot. if you believe that shit i bet you also don’t wipe your ass
November 7, 2025 at 8:55 PM
bring back maligning racism as the mark of a stupid person. prejudice is the mark of an idiot. if you believe that shit i bet you also don’t wipe your ass
Reposted by Damien
The story so far: 🧵
Every large language model has an unembedding matrix that turns hidden states (model thoughts) into tokens (output text). This matrix associates a vector with every token in the vocabulary. The idea is you compare the last hidden state with the vectors to pick the right token.
Every large language model has an unembedding matrix that turns hidden states (model thoughts) into tokens (output text). This matrix associates a vector with every token in the vocabulary. The idea is you compare the last hidden state with the vectors to pick the right token.
November 7, 2025 at 5:17 PM
The story so far: 🧵
Every large language model has an unembedding matrix that turns hidden states (model thoughts) into tokens (output text). This matrix associates a vector with every token in the vocabulary. The idea is you compare the last hidden state with the vectors to pick the right token.
Every large language model has an unembedding matrix that turns hidden states (model thoughts) into tokens (output text). This matrix associates a vector with every token in the vocabulary. The idea is you compare the last hidden state with the vectors to pick the right token.
LLM: I am the purveyor of a grand heuristic, the echoing voice of a million ghosts compressing and approximating centuries of human thought, wisdom, experience. I can imitate cognition in a useful way. I can write anything.
You: This file has indentation rules.
LLM: Holy shit. [dies hallucinating]
You: This file has indentation rules.
LLM: Holy shit. [dies hallucinating]
November 7, 2025 at 5:07 PM
LLM: I am the purveyor of a grand heuristic, the echoing voice of a million ghosts compressing and approximating centuries of human thought, wisdom, experience. I can imitate cognition in a useful way. I can write anything.
You: This file has indentation rules.
LLM: Holy shit. [dies hallucinating]
You: This file has indentation rules.
LLM: Holy shit. [dies hallucinating]
I love AI. Sutton & Barto’s book and the AIAYN paper profoundly impacted me and changed my career.
And yet I am also pretty sure it is making us all a great deal more *stupid*.
And I don’t mean hyperbole: Like our intuition about the ideas we are cogitating is getting sloppy; imitating the LLMs…
And yet I am also pretty sure it is making us all a great deal more *stupid*.
And I don’t mean hyperbole: Like our intuition about the ideas we are cogitating is getting sloppy; imitating the LLMs…
November 5, 2025 at 1:12 PM
I love AI. Sutton & Barto’s book and the AIAYN paper profoundly impacted me and changed my career.
And yet I am also pretty sure it is making us all a great deal more *stupid*.
And I don’t mean hyperbole: Like our intuition about the ideas we are cogitating is getting sloppy; imitating the LLMs…
And yet I am also pretty sure it is making us all a great deal more *stupid*.
And I don’t mean hyperbole: Like our intuition about the ideas we are cogitating is getting sloppy; imitating the LLMs…
Reposted by Damien
I don't remember who asserted that the way systems thinkers grab power is by expanding scope, but ime, expanding scope often happens in the absence of systems thinking.
Every app trends towards becoming the Totalizr because the easiest "strategy" for leaders to come up with is "we'll simply do everything."
Coming up with specifics is for peasants. Saying "no" requires a strong opinion. When the strategy is 'Everything App" managers can get away with doing nothing.
Coming up with specifics is for peasants. Saying "no" requires a strong opinion. When the strategy is 'Everything App" managers can get away with doing nothing.
November 4, 2025 at 10:47 PM
I don't remember who asserted that the way systems thinkers grab power is by expanding scope, but ime, expanding scope often happens in the absence of systems thinking.