Gordon Weakliem
@2hardproblems.bsky.social
Software and other stuff
"There are two hard problems in CS: naming things, cache expiration, and off by one errors"
http://eighty-twenty.net
"There are two hard problems in CS: naming things, cache expiration, and off by one errors"
http://eighty-twenty.net
Cool, so if you're wealthy enough to afford a private jet, you're not affected by the ATC issues.
www.cnbc.com/2025/11/10/p...
www.cnbc.com/2025/11/10/p...
Private-jet demand is on the rise amid government shutdown, says Flexjet CEO
The FAA plans to curb private-jet traffic at 12 major U.S. airports starting Monday.
www.cnbc.com
November 10, 2025 at 4:47 PM
Cool, so if you're wealthy enough to afford a private jet, you're not affected by the ATC issues.
www.cnbc.com/2025/11/10/p...
www.cnbc.com/2025/11/10/p...
Reposted by Gordon Weakliem
This sounds cynical—but represents a huge advance over empty “AGI” speculation.
It’s a political question, not a technical one. Without social equality models *cannot do* many kinds of work (eg, negotiate agreements or manage workers). So they will only be human-equivalent if we decide they are.
It’s a political question, not a technical one. Without social equality models *cannot do* many kinds of work (eg, negotiate agreements or manage workers). So they will only be human-equivalent if we decide they are.
intelligence is the thing which i have. admitting things are intelligent means considering them morally and socially equal to me. i will never consider a computer morally or socially equal to me. therefore no computer program will ever be intelligent
November 2, 2025 at 11:01 AM
This sounds cynical—but represents a huge advance over empty “AGI” speculation.
It’s a political question, not a technical one. Without social equality models *cannot do* many kinds of work (eg, negotiate agreements or manage workers). So they will only be human-equivalent if we decide they are.
It’s a political question, not a technical one. Without social equality models *cannot do* many kinds of work (eg, negotiate agreements or manage workers). So they will only be human-equivalent if we decide they are.
One analogy that worth thinking about: what is it like to live in a world where some people have superpowers?
October 23, 2025 at 1:02 AM
One analogy that worth thinking about: what is it like to live in a world where some people have superpowers?
"you should think of [an agent] almost like an employee or an intern that you would hire to work with you"
I completely agree with this, and I always try to frame working with an agent this way. The bit about animals vs ghosts is a really interesting analogy.
www.dwarkesh.com/p/andrej-kar...
I completely agree with this, and I always try to frame working with an agent this way. The bit about animals vs ghosts is a really interesting analogy.
www.dwarkesh.com/p/andrej-kar...
Andrej Karpathy — AGI is still a decade away
"The problems are tractable, but they're still difficult”
www.dwarkesh.com
October 22, 2025 at 10:41 PM
"you should think of [an agent] almost like an employee or an intern that you would hire to work with you"
I completely agree with this, and I always try to frame working with an agent this way. The bit about animals vs ghosts is a really interesting analogy.
www.dwarkesh.com/p/andrej-kar...
I completely agree with this, and I always try to frame working with an agent this way. The bit about animals vs ghosts is a really interesting analogy.
www.dwarkesh.com/p/andrej-kar...
“It is hard to know exactly what will happen to an industry build on an apprenticeship model when it no longer needs apprentices”
This newsletter is the result of a “screwup” caused by Money Stuff's “antiquated” computer system. www.bloomberg.com/opinion/news...
OpenAI Is Building a Banker
Also abstract copper, 777 Partners, in-kind Bitcoin ETFs and the AWS outage.
www.bloomberg.com
October 21, 2025 at 7:29 PM
“It is hard to know exactly what will happen to an industry build on an apprenticeship model when it no longer needs apprentices”
If you’re scared of agents, what do you do as an EM or PM faced with the output of a high performing team. Isn’t it the same problem?
Andrej Karpathy’s take on AI coding agents feels grounded. The industry’s chasing full autonomy when models still hallucinate too much.
Agents that churn out a thousand lines of code leave you either blindly trusting them or slogging through reviews. These tools should embrace their fallibility.
Agents that churn out a thousand lines of code leave you either blindly trusting them or slogging through reviews. These tools should embrace their fallibility.
October 19, 2025 at 11:37 PM
If you’re scared of agents, what do you do as an EM or PM faced with the output of a high performing team. Isn’t it the same problem?
Reposted by Gordon Weakliem
it's kinda weird that all the software i am expected to use for work are all written by distributed teams, go, python, postgres, linux, chrome, k8s etc
and despite being told "the best teams work in an office together" i don't know of any software i use that's actually written that way
and despite being told "the best teams work in an office together" i don't know of any software i use that's actually written that way
October 13, 2025 at 5:00 PM
it's kinda weird that all the software i am expected to use for work are all written by distributed teams, go, python, postgres, linux, chrome, k8s etc
and despite being told "the best teams work in an office together" i don't know of any software i use that's actually written that way
and despite being told "the best teams work in an office together" i don't know of any software i use that's actually written that way
After all the shit happened in The Odyssey, Odysseus put an oar over his shoulder and marched inland until people asked what he was carrying & he knew he'd found where he belonged.
I just gave a project proposal to a prospective client organized into sprints. She asked what a sprint was.
I just gave a project proposal to a prospective client organized into sprints. She asked what a sprint was.
October 3, 2025 at 7:58 PM
After all the shit happened in The Odyssey, Odysseus put an oar over his shoulder and marched inland until people asked what he was carrying & he knew he'd found where he belonged.
I just gave a project proposal to a prospective client organized into sprints. She asked what a sprint was.
I just gave a project proposal to a prospective client organized into sprints. She asked what a sprint was.
People focus on the asbestos quote, and corporate America feels extremely broken. I like the closing: "It's a grab-bag of useful (sometimes very useful) tools that can sometimes make workers' lives better, when workers get to decide how and when they're used."
pluralistic.net/2025/09/27/e...
pluralistic.net/2025/09/27/e...
Pluralistic: The real (economic) AI apocalypse is nigh (27 Sep 2025) – Pluralistic: Daily links from Cory Doctorow
pluralistic.net
September 30, 2025 at 12:56 PM
People focus on the asbestos quote, and corporate America feels extremely broken. I like the closing: "It's a grab-bag of useful (sometimes very useful) tools that can sometimes make workers' lives better, when workers get to decide how and when they're used."
pluralistic.net/2025/09/27/e...
pluralistic.net/2025/09/27/e...
Reposted by Gordon Weakliem
The paperback is ready!!!! Great Power. Questionable responsibility. When the capes are gone the burden remains. #readerscommunity #superheroes #shortstory #indieauthors
www.amazon.com/dp/B0FSZ5QYR...
www.amazon.com/dp/B0FSZ5QYR...
SUPERPOWERLESS
SUPERPOWERLESS [Salavon, Ian] on Amazon.com. *FREE* shipping on qualifying offers. SUPERPOWERLESS
www.amazon.com
September 28, 2025 at 12:02 PM
The paperback is ready!!!! Great Power. Questionable responsibility. When the capes are gone the burden remains. #readerscommunity #superheroes #shortstory #indieauthors
www.amazon.com/dp/B0FSZ5QYR...
www.amazon.com/dp/B0FSZ5QYR...
Reposted by Gordon Weakliem
Meanwhile, I see deep questions on the AI side that should definitely warrant some humanities input. And I don't mean the usual ethics corner, but: how to design reasoning structure/flow, personality tuning (what is the "I" in the model?), rethinking tokenization. Won't happen for some time.
Keeping “digital humanities” together as a field is going to get so hard in the era of AI. The shear forces are intense: I see lots of humanities profs who were DH-friendly reposting Zitron.
And while E.Z. is a Limbaugh-tier thinker imo, they’re not wrong that AI is disrupting things they value. +
And while E.Z. is a Limbaugh-tier thinker imo, they’re not wrong that AI is disrupting things they value. +
September 4, 2025 at 11:28 PM
Meanwhile, I see deep questions on the AI side that should definitely warrant some humanities input. And I don't mean the usual ethics corner, but: how to design reasoning structure/flow, personality tuning (what is the "I" in the model?), rethinking tokenization. Won't happen for some time.
Maybe instead of forbidding using an LLM during a coding interview, we should make candidates use one that appears helpful but gives wrong answers, to test how a candidate's BS detector works.
September 2, 2025 at 8:02 PM
Maybe instead of forbidding using an LLM during a coding interview, we should make candidates use one that appears helpful but gives wrong answers, to test how a candidate's BS detector works.
Reading old performance reviews and came across a line - “Dare thinks being correct is more important than convincing people” which at the time struck me as absurd as what’s more important than being right?
Over time I learned if you can’t change what work happens then being correct doesn’t matter.
Over time I learned if you can’t change what work happens then being correct doesn’t matter.
September 1, 2025 at 3:59 PM
I don't like the "make everything a chatbot" trend. Had to reset a password yesterday, got directed into a chatbot flow that would have been better handled as a plain old web form, and even worse because I had to go check email in between to click on reset links anyway.
August 27, 2025 at 5:16 PM
I don't like the "make everything a chatbot" trend. Had to reset a password yesterday, got directed into a chatbot flow that would have been better handled as a plain old web form, and even worse because I had to go check email in between to click on reset links anyway.
Seems like Next.js is becoming the low-friction deployment option - Vercel is heavily optimized around this, AWS Amplify v2 now has that as a default. A little bit of struggle to find something that will host a simple FastAPI application without reading a few articles about the "yes, but" issues
August 26, 2025 at 12:52 PM
Seems like Next.js is becoming the low-friction deployment option - Vercel is heavily optimized around this, AWS Amplify v2 now has that as a default. A little bit of struggle to find something that will host a simple FastAPI application without reading a few articles about the "yes, but" issues
One of my favorite tech quotes: “people will do the stupidest things, and have excellent reasons for doing them”
(I think that was Mike Swaine writing in Dr. Dobbs Journal)
(I think that was Mike Swaine writing in Dr. Dobbs Journal)
August 24, 2025 at 2:45 PM
One of my favorite tech quotes: “people will do the stupidest things, and have excellent reasons for doing them”
(I think that was Mike Swaine writing in Dr. Dobbs Journal)
(I think that was Mike Swaine writing in Dr. Dobbs Journal)
I’ve been watching the Netflix documentary on Jerry Jones and the Cowboys. I lived in DFW at the time, so you couldn’t escape it. My favorite part is seeing all the off field stuff and remembering what life was like back then.
August 24, 2025 at 2:29 PM
I’ve been watching the Netflix documentary on Jerry Jones and the Cowboys. I lived in DFW at the time, so you couldn’t escape it. My favorite part is seeing all the off field stuff and remembering what life was like back then.
Good news! If you bought $CSCO in May 2000, you’re in the black!
August 22, 2025 at 10:44 AM
Good news! If you bought $CSCO in May 2000, you’re in the black!
Recently I looked on X and saw a thread that showed how to train the new Gemma 270m model. Except it left out a few critical details so it didn’t run. I dug deeper and the poster had copied the entire thing from an Unsloth demo. That’s X, literal den of thieves. I spend a lot more time on here.
August 21, 2025 at 12:19 PM
Recently I looked on X and saw a thread that showed how to train the new Gemma 270m model. Except it left out a few critical details so it didn’t run. I dug deeper and the poster had copied the entire thing from an Unsloth demo. That’s X, literal den of thieves. I spend a lot more time on here.
"Hey, a carrot can be a stick if you hit someone hard enough with it"
August 19, 2025 at 5:13 PM
"Hey, a carrot can be a stick if you hit someone hard enough with it"
I got to play around with gemma-3-270m this weekend, I managed to do a couple of fine tunes on it using Colab free tier. This is pretty exciting!
Unsloth has a great tutorial: colab.research.google.com/github/unslo...
Unsloth has a great tutorial: colab.research.google.com/github/unslo...
Google Colab
colab.research.google.com
August 18, 2025 at 2:39 AM
I got to play around with gemma-3-270m this weekend, I managed to do a couple of fine tunes on it using Colab free tier. This is pretty exciting!
Unsloth has a great tutorial: colab.research.google.com/github/unslo...
Unsloth has a great tutorial: colab.research.google.com/github/unslo...
Reposted by Gordon Weakliem
The only certainty is rich men trying to avoid death and taxes.
August 13, 2025 at 10:43 PM
The only certainty is rich men trying to avoid death and taxes.
This really is better than Discover. I’m very curious to know how it was built
Welcome to the ✨For You✨ feed!
It finds people who liked the same posts as you, and shows you what else they've liked recently.
📌 Pin to add it to your top bar
❤️ Like the feed and repost to spread the goodness
It finds people who liked the same posts as you, and shows you what else they've liked recently.
📌 Pin to add it to your top bar
❤️ Like the feed and repost to spread the goodness
August 13, 2025 at 12:00 PM
This really is better than Discover. I’m very curious to know how it was built