Alex Gude
@alexgude.com
Machine learning engineer in Silicon Valley! Formerly a particle physicist at CERN and alumni of Insight. He/Him
https://alexgude.com
@[email protected]
https://alexgude.com
@[email protected]
Pinned
Alex Gude
@alexgude.com
· Nov 26
The Darfsteller
The Darfsteller, by Walter M. Miller Jr., is a Hugo Award-winning novelette about the obsolescence of the human artist. It follows Ryan Thornier, a former stage idol reduced to working as a janitor in...
alexgude.com
Seventy years ago Walter M Miller Jr wrote how creative work would be the first thing we automate. The Darfsteller follows an actor replaced by robots, which are both cheaper and preferred by the audience.
It's more relevant than ever in the age of Gen AI.
alexgude.com/books/the_da...
#BookReview
It's more relevant than ever in the age of Gen AI.
alexgude.com/books/the_da...
#BookReview
I absolutely believe that environmental effects dominate. I'm more than a 10x engineer at my new company compared to me at my old one.
And it's entirely because we've structured the organization and tech stack to allow it. I'm mostly the same dev I was before, just unchained.
And it's entirely because we've structured the organization and tech stack to allow it. I'm mostly the same dev I was before, just unchained.
We saw no evidence for these super devs looking across cycle times in 216 orgs and over 55k cycle times over a year of software development work. I think consistent individual velocity effects are present and possible but honestly, environment, task variance and compounding other factors mute them.
December 30, 2025 at 4:34 PM
I absolutely believe that environmental effects dominate. I'm more than a 10x engineer at my new company compared to me at my old one.
And it's entirely because we've structured the organization and tech stack to allow it. I'm mostly the same dev I was before, just unchained.
And it's entirely because we've structured the organization and tech stack to allow it. I'm mostly the same dev I was before, just unchained.
Reposted by Alex Gude
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
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.
Reposted by Alex Gude
Technically they buy one copy exactly for the Machine Dimension and then are like “thanks! 😊 for the book! 😊😊😊 we all read it!!!” and when humans are like “can I get more than one copy’s worth of royalties” they’re like “sorry! 😊 we’re going through a tunnel right now!!!”
December 29, 2025 at 5:29 AM
Technically they buy one copy exactly for the Machine Dimension and then are like “thanks! 😊 for the book! 😊😊😊 we all read it!!!” and when humans are like “can I get more than one copy’s worth of royalties” they’re like “sorry! 😊 we’re going through a tunnel right now!!!”
Reposted by Alex Gude
Every poster who is all in on AI being a supply-driven fad should be made to explain how ChatGPT is about to hit a billion weekly active users.
I worked on 3DTV at Discovery. I followed VR very closely for several years. I know a supply-driven phenomenon when I see one. This ain’t that.
I worked on 3DTV at Discovery. I followed VR very closely for several years. I know a supply-driven phenomenon when I see one. This ain’t that.
December 28, 2025 at 1:08 AM
Every poster who is all in on AI being a supply-driven fad should be made to explain how ChatGPT is about to hit a billion weekly active users.
I worked on 3DTV at Discovery. I followed VR very closely for several years. I know a supply-driven phenomenon when I see one. This ain’t that.
I worked on 3DTV at Discovery. I followed VR very closely for several years. I know a supply-driven phenomenon when I see one. This ain’t that.
Reposted by Alex Gude
Andrej Karpathy is worried about keeping up with software engineering practices
December 27, 2025 at 9:40 AM
Andrej Karpathy is worried about keeping up with software engineering practices
Reposted by Alex Gude
Academics literally cannot make Gen AI go away. As in it is not possible to make it happen on a political, legal, and technical level. We do not and will not command the massive violence necessary to make chatGPT or Claude disappear. Responsible pedagogy grapples with the fact!
No one says to legit business owners, "hey I see there's an underground market selling toxic knock-offs of your product: how will you incorporate the toxic product AND protect your customers against its harms?" But apparently it's fine to make this suggestion to educators with respect to AI.
December 26, 2025 at 8:04 AM
Academics literally cannot make Gen AI go away. As in it is not possible to make it happen on a political, legal, and technical level. We do not and will not command the massive violence necessary to make chatGPT or Claude disappear. Responsible pedagogy grapples with the fact!
Come on, Dems. be honest. Deep down, isn't there at least a tiny seedling-sized trace of testosterone left in your brain that looks at this BATTLESHIP video and thinks:
"...okay, yeah, that's kinda awesome"
Be honest...
"...okay, yeah, that's kinda awesome"
Be honest...
December 24, 2025 at 11:19 PM
Come on, Dems. be honest. Deep down, isn't there at least a tiny seedling-sized trace of testosterone left in your brain that looks at this BATTLESHIP video and thinks:
"...okay, yeah, that's kinda awesome"
Be honest...
"...okay, yeah, that's kinda awesome"
Be honest...
Reposted by Alex Gude
Good news, someone ripped my video and put it on the Internet Archive. If you're having trouble downloading from Filemail, here it is:
archive.org/details/60mi...
archive.org/details/60mi...
Pulled 60 Minutes segment on CECOT : CBS : Free Download, Borrow, and Streaming : Internet Archive
This is a screen recording of a 60 Minutes segment about the Centro de Confinamiento del Terrorismo (CECOT) prison in El Salvador, which was intended to be...
archive.org
December 23, 2025 at 12:03 AM
Good news, someone ripped my video and put it on the Internet Archive. If you're having trouble downloading from Filemail, here it is:
archive.org/details/60mi...
archive.org/details/60mi...
Reposted by Alex Gude
van Rooij says she's not a creationist and is instead providing this "article" that has evolution in the title.
it is INDISTINGUISHABLE from intelligent design arguments around complexity.
you can dislike ai but don't fall for creationist cranks.
onlinelibrary.wiley.com/doi/full/10....
it is INDISTINGUISHABLE from intelligent design arguments around complexity.
you can dislike ai but don't fall for creationist cranks.
onlinelibrary.wiley.com/doi/full/10....
December 23, 2025 at 1:48 AM
van Rooij says she's not a creationist and is instead providing this "article" that has evolution in the title.
it is INDISTINGUISHABLE from intelligent design arguments around complexity.
you can dislike ai but don't fall for creationist cranks.
onlinelibrary.wiley.com/doi/full/10....
it is INDISTINGUISHABLE from intelligent design arguments around complexity.
you can dislike ai but don't fall for creationist cranks.
onlinelibrary.wiley.com/doi/full/10....
Reposted by Alex Gude
Reposted by Alex Gude
this twitter thread is van rooji explaining a paper doing the same complexity theory analysis to disprove that evolution by natural selection can ever result in rational behavior, back in 2021 x.com/IrisVanRooij...
Iris van Rooij 💭 on X: "🚨 Recently assigned to an issue. Good opportunity to redo a thread on it! "Naturalism, tractability and the adaptive toolbox" (Synthese, 2021), by Patricia Rich, @MarkBlokpoel, Ronald de Haan, Maria Otworowska, Marieke Sweers, Todd Wareham, @IrisVanRooij https://t.co/tRTHwyZJgf" / X
🚨 Recently assigned to an issue. Good opportunity to redo a thread on it! "Naturalism, tractability and the adaptive toolbox" (Synthese, 2021), by Patricia Rich, @MarkBlokpoel, Ronald de Haan, Maria Otworowska, Marieke Sweers, Todd Wareham, @IrisVanRooij https://t.co/tRTHwyZJgf
x.com
December 22, 2025 at 7:17 AM
this twitter thread is van rooji explaining a paper doing the same complexity theory analysis to disprove that evolution by natural selection can ever result in rational behavior, back in 2021 x.com/IrisVanRooij...
Reposted by Alex Gude
One thing I love is obsolete technology being encoded in our language. Words like “footage” and “rewind” and “dial (as a phone)” that literally describe a physicality that doesn’t exist anymore, yet lingers linguistically. Anybody have any other examples?
December 22, 2025 at 1:49 PM
One thing I love is obsolete technology being encoded in our language. Words like “footage” and “rewind” and “dial (as a phone)” that literally describe a physicality that doesn’t exist anymore, yet lingers linguistically. Anybody have any other examples?
The great part about being a "shut up an calculate" scientist is you can ignore papers that say things shouldn't work in theory, which do work in practice.
December 21, 2025 at 9:23 PM
The great part about being a "shut up an calculate" scientist is you can ignore papers that say things shouldn't work in theory, which do work in practice.
Reposted by Alex Gude
ML is a subset of AI, and AI is a subset of ML. But the second AI isn’t the same thing as the first AI (except it is partially because it’s a subset of itself)
December 21, 2025 at 8:01 PM
ML is a subset of AI, and AI is a subset of ML. But the second AI isn’t the same thing as the first AI (except it is partially because it’s a subset of itself)
Reposted by Alex Gude
One of my professors in undergrad was a “cognitive scientist” who would insist he had proved P=NP because a certain minimal surfaces problem was NP-hard but you could use soap films to solve it in O(1).
That represented a substantially more rigorous and defensible argument than whatever this is:
That represented a substantially more rigorous and defensible argument than whatever this is:
We know — from first principles — that LLMs can only create AI slop.
Why do people keep needing empirical tests?
Why do people keep needing empirical tests?
The intractability proof (a.k.a. Ingenia theorem) implies that any attempts to scale up AI-by-Learning to situations of real-world, human-level complexity will consume an astronomical amount of resources (see Box 1 for an explanation). 13/n
December 21, 2025 at 5:13 PM
One of my professors in undergrad was a “cognitive scientist” who would insist he had proved P=NP because a certain minimal surfaces problem was NP-hard but you could use soap films to solve it in O(1).
That represented a substantially more rigorous and defensible argument than whatever this is:
That represented a substantially more rigorous and defensible argument than whatever this is:
I don't have as much structure built around Gemini CLI as I do Claude Code, so this isn't a fair test, but...
Intern Gemini is probably not getting a return offer, where as Intern Claude definitely is.
Intern Gemini is probably not getting a return offer, where as Intern Claude definitely is.
Really having to sit the (LLM) intern down today and say:
Yeah, look, I know having a really cool and smart library is fun to write, but we should write the dumbest thing possible instead. How about that?
Yeah, look, I know having a really cool and smart library is fun to write, but we should write the dumbest thing possible instead. How about that?
December 21, 2025 at 7:36 PM
I don't have as much structure built around Gemini CLI as I do Claude Code, so this isn't a fair test, but...
Intern Gemini is probably not getting a return offer, where as Intern Claude definitely is.
Intern Gemini is probably not getting a return offer, where as Intern Claude definitely is.
Reposted by Alex Gude
People talking about "LLMs can't make new information" is so fucking funny. Ah yeah, AlphaZero hasn't found new chess strategies. Protein folding models don't make information.
December 21, 2025 at 4:20 PM
People talking about "LLMs can't make new information" is so fucking funny. Ah yeah, AlphaZero hasn't found new chess strategies. Protein folding models don't make information.
Really having to sit the (LLM) intern down today and say:
Yeah, look, I know having a really cool and smart library is fun to write, but we should write the dumbest thing possible instead. How about that?
Yeah, look, I know having a really cool and smart library is fun to write, but we should write the dumbest thing possible instead. How about that?
December 21, 2025 at 5:50 PM
Really having to sit the (LLM) intern down today and say:
Yeah, look, I know having a really cool and smart library is fun to write, but we should write the dumbest thing possible instead. How about that?
Yeah, look, I know having a really cool and smart library is fun to write, but we should write the dumbest thing possible instead. How about that?
Reposted by Alex Gude
This Is How You Lose the Time War was a quick, fun read that did a great job of making me feel, but sort of vanished from my head as soon as I shut it.
alexgude.com/books/this_i...
#BookReview
alexgude.com/books/this_i...
#BookReview
This Is How You Lose the Time War
This Is How You Lose the Time War, by Amal El-Mohtar and Max Gladstone, is a story about two agents—Red and Blue—working for opposite factions from the far future, changing the past in an attempt to w...
alexgude.com
December 20, 2025 at 6:35 PM
This Is How You Lose the Time War was a quick, fun read that did a great job of making me feel, but sort of vanished from my head as soon as I shut it.
alexgude.com/books/this_i...
#BookReview
alexgude.com/books/this_i...
#BookReview
This is *EXACTLY* what I was getting at, but better expressed, with my forum member post.
These things are technologically miraculous.
These things are technologically miraculous.
THERE'S A MACHINE THAT CAN EXPLAIN NEW THEOREMS IN NATURAL LANGUAGE WHAT IS WRONG WITH YOU PEOPLE. If someone showed you it in 2010 you would literally assume it was faked
December 19, 2025 at 6:28 PM
This is *EXACTLY* what I was getting at, but better expressed, with my forum member post.
These things are technologically miraculous.
These things are technologically miraculous.
Reposted by Alex Gude
THERE'S A MACHINE THAT CAN EXPLAIN NEW THEOREMS IN NATURAL LANGUAGE WHAT IS WRONG WITH YOU PEOPLE. If someone showed you it in 2010 you would literally assume it was faked
December 19, 2025 at 5:40 AM
THERE'S A MACHINE THAT CAN EXPLAIN NEW THEOREMS IN NATURAL LANGUAGE WHAT IS WRONG WITH YOU PEOPLE. If someone showed you it in 2010 you would literally assume it was faked
Reposted by Alex Gude
Not only are most commercial AI systems also search engines, they are orders of magnitude better than googles standard search engine.
chatgpt is literally a search engine and has been for ages tho lol
day 473 of having to explain to coworkers that ChatGPT is not a search engine
these people aren't even boomers they're engineers in their thirties and forties, it's so over
these people aren't even boomers they're engineers in their thirties and forties, it's so over
December 18, 2025 at 12:00 AM
Not only are most commercial AI systems also search engines, they are orders of magnitude better than googles standard search engine.