QIP: Quick Isolated Portable WebAssembly components: https://xitter.cf/t.co/fhKA3m9FXV blog: https://xitter.cf/t.co/EtrtGaXYON

Melbourne, Australia
Joined August 2013
Patrick Smith retweeted
LLMs make it easier than ever for normal software to get down to zero allocations.
3
2
58
4,400
Patrick Smith retweeted
Got a cheap embroidery machine off amazon for $300 where the biggest complaint is the mandatory $12/mo app that you’re forced to use. Had ChatGPT create its own app that communicates with the machine. Generative art patches incoming!
124
128
44
4,550
186,046
Patrick Smith retweeted
Corrupt autocracies prefer to do business with other corrupt autocracies because the profits are personal on both sides, not for the state or the country. No oversight, no investigations.
The US is willing to betray its closest allies in Europe, a market of 450 million people and an economy worth nearly $20 trillion, just to cut shady, corrupt deals with Putin's Russia, an economy roughly the size of Italy's. What an absolutely insane foreign policy.
63
1,760
29
5,901
67,943
Patrick Smith retweeted
Yesterday, someone posted a task runner claimed to be 39x faster than Turborepo. I got really excited, thinking they found some edge case in Turborepo and that I could make something way faster. I open the benchmark repo, and its just straight up, flat out incorrect. Like, not even sorta useful. Not only was it not measuring the task runner parts of the script, it was overfit to a highly specific, completely unrealistic workload. 💀 The entire project is obviously Opus 5.5 vibe coded and Opus cheated the goal. Easy to miss if you're not experienced in build systems. Forgivable! So, I open a PR to show that the algorithm being used can also make scheduling slower, by any arbitrary factor, with only a tiny adjustment. I chose 100x slower than Turborepo for demonstrative purposes. The author fixed some of the problems but there are still grave issues. They're still comparing on an overfit bench that doesn't hold up when you insert the slightest variable, and *insisting* that its right. It is very obviously incorrect, but oh, well. After revision, Turborepo is only 1.36x slower. Down from 39x, but a more realistic number nonetheless. So I figure: "Great! There must be something I can learn to make Turborepo faster." I go digging, and it turns out the other tool is only faster still because it has correctness bugs. Once they correct those, they'll be roughly on par with Turborepo. On top of that, the numbers they post now report with a mix of "% faster" and "✗ faster", using "% faster" whenever the difference isn't big enough to make their tool look good. 1.36x gets turned into "36%", which looks like more than it is to a casual observer. This is a story of how I thought I was going to speed up Turborepo by 39x, and got nothing out of it. I'm not mad; I'm just disappointed. 😄
11
7
8
331
29,948
Patrick Smith retweeted
43
2,365
75
30,682
283,778
Patrick Smith retweeted
Actual expertise is the most important thing right now in tech. Knowing how things work, and how they are implemented on a deep level is more important than ever.
13
26
5
380
5,920
Patrick Smith retweeted
25 years ago, on October 22, 2001, Grand Theft Auto III was released. This is the team that made it. (I'm second from the right) Here's what I remember about making it. For most of development, GTA3 was in a sorry state. It crashed frequently, the frame rate was terrible and the map was full of holes. Only in the last six months did everything come together. For the longest time, it looked like the frame rate was going to be 20 fps. Only in the final months did we manage to hit 30. In 1999, DMA Design, based in Dundee, Scotland, was going through some changes. The company had recently been bought by Take-Two Interactive, and a satellite office had opened in Edinburgh. Space Station Silicon Valley for the N64 had been finished in 1998, but there was no new project lined up. Leslie Benzies and Aaron Garbut started working on GTA3. Unprompted. Meanwhile, the original GTA1 and GTA2 team had started work on GTA2.5, set in Miami. It was called 2.5 because it used the 2.5D (isometric) perspective. In March 2000, the Dundee office was closed and GTA2.5 was cancelled. Take-Two/Rockstar gave GTA3 the green light. Everything was written from scratch. Only two people worked on both GTA2 and GTA3, and no code or art from GTA2 was reused. The way we worked was organic. People would come up with ideas for missions and we'd just try them. If they were fun, they were in. The story was secondary. In fact, the opening plot involving Catalina trying to kill Claude was added well into development. Leslie Benzies (the producer) plastered a wall in the boardroom with Post-it notes. They were yellow, green and pink, representing missions, contacts and locations. Missions were moved around all the time, often handed from one contact to another. Only once the voice acting was recorded were these things set in stone. You sometimes come across locations on the map that are more detailed than you'd expect. Chances are there used to be a mission contact there that got moved elsewhere. Working on GTA3 was a lot of fun. Every day there was something new going in. A new mission, radio station, car, pedestrian, building or whatever. The game was growing at an incredible rate. Other than the multiplayer mode we did some work on, I can’t remember a lot of wastage. Although everybody worked hard, I don't remember many people doing crazy hours. On the rare occasions I was in over the weekend, there were usually only one or two others. Even during crunch time, most people left the office around 9 or 10 in the evening. I always thought it was better to work eight focused hours than spend your life in the office. Although we liked the game, it wasn't at all clear that it would be a hit. GTA3 had a disappointing showing at E3 in May 2001. Most people weren't interested, and those who were thought it was too violent. It had an OK launch, but sales never tapered off. Our office was in Leith, Edinburgh. It was a fairly run-down neighbourhood at the time. Just a couple of years before we arrived, J.K. Rowling had been writing the first Harry Potter in that very same neighbourhood. The thing I remember most is the camaraderie. In a small team, everybody relies on each other and everybody has a lot of input. Most of us were young and had only recently moved to Edinburgh. We went to the pub together. It felt like a group of friends who happened to be working on a project together. xitter.cf/ObbeVermeij/status/178… xitter.cf/ObbeVermeij/status/175… xitter.cf/ObbeVermeij/status/187… xitter.cf/ObbeVermeij/status/182… xitter.cf/ObbeVermeij/status/179…
For gta3, Vice and SA; Memory on the ps2 was tight. We had to limit the number of used vehicle models to 8. My code would occasionally pick a car model to be phased out. Once there were none left on the map, this model would be removed and a new model could be loaded. The code would pick a model that was appropriate for the area (sports cars in business district, old cars in run-down areas etc). Most of the time this worked fine. A few things could throw a spanner in the works: If the player had gone up the wanted levels, some of the 8 slots would be taken up by police cars, FIB and swat vans and the police heli. Level designers would sometimes load specific car models for missions. Often these were not appropriate to be used for ambient traffic. Ambulance and firetrucks could be needed if there was an accident. If the player travelled to a different area of the map; some of the loaded cars would look out of place and could not be used. It would take a while for replacement models to be loaded. The game would keep the last player driven car, as it would be jarring if the player went back and it wasn’t there. At times there would only be 1 or 2 cars models available for ambient traffic. This lack of variation was particularly noticeable if the player was driving the one car that had to be used. To alleviate the problem, garages remove cars inside of them once the door has closed. This way the model could be removed. When the player opened the door, the cars were re-created in the same spot. Npc models were loaded/removed the same way. It just wasn’t as noticeable if variety was lacking. For each game we picked a generic looking character model that looked appropriate in each area of the map. This model was never removed and always ready to fill in. Screenshot from this vid: youtube.com/watch?v=KWhjv4J6…
273
1,180
47
13,433
404,256
Patrick Smith retweeted
1/3 Musk ordered last week's slowdown of Starshield, which Ukraine fights on, after talking with Trump, per @WSJ. In '23 the Pentagon signed a contract with SpaceX to “prevent Musk from turning off the switch," per fmr Under Sec of Defense Colin Kahl. newyorker.com/magazine/2023/…
186
2,856
144
7,273
476,858
Patrick Smith retweeted
Theres a peril in reading only the diffs that an agent produces. It's (nearly) a given that the change is going to be locally valid with itself. Your job is to gauge its coherence with the rest of the code that you are quickly losing touch with. Best to open those files up, read the surrounding and called code, and let it soak for a bit.
2
1
10
367
Patrick Smith retweeted
“we will pay any price….to assure the success and survival of liberty.” Is this still true? Do we prize liberty or do we prize power and purse? No more clear test than the choice between Ukraine and Russia. Abandon Ukraine and we abandon liberty. Eventually even our own.
3,110
5,923
427
29,298
906,462
Patrick Smith retweeted
Replying to @SpatiallyMe
This is why I am excited for the Duo, it seems to be inspiring developers to make apps that bring back the whimsy, like the days when the App Store first launched.
1
1
2
44
3,148
Patrick Smith retweeted
I am so much more exhausted when I program with AI than without it. I reflected a bit, and I think I know why. First, I think about how to solve a problem. This step is the same with and without AI. Then, with AI I have to write a prompt with a decent bit of precision and somewhat more long-term than if I just jumped into straight implementing it and figuring out the details as I go. Then you start up the agent, and you immediately have some other work to do and you jump to another prompt for another task, where you also think really hard and try to formulate what you want quite well. By the time you're done, first agent has some code for you to review, and you can't slack - you need to actually review it and be diligent about it. And when you're done - second task is there for your review. So really there's not time to rest. When I program manually, I can decompress while typing the code in, I can rely on my muscle memory for typing and navigating the codebase, so there's periods of deep think and some relaxation. With AI, there's only deep think. And as soon as you try to think less - you open the doors for slop. It's so tiring.
101
48
14
948
28,829
Patrick Smith retweeted
Replying to @Paul_Kinlan
Agree. Claude was able to do this better than I could since Opus 4.8 or so. 5.5 models are now very good at it. All you do is give the model 1. a goal (eg. “Make p95 ci time <10 mins”) 2. a way to measure its result (eg. “Read timings in bigquery after each change”) 3. a sense of how hard to try (eg. “Don’t stop till it’s done, it might take a few weeks of experimenting”) This pretty much works for most things. I often also ask the model to show me a chart with progress to goal every day, or to dm me a live artifact with the result
13
25
6
293
49,996
Patrick Smith retweeted
How long should a micro benchmark run? matklad.github.io/2026/10/05…
1
6
49
2,154
Patrick Smith retweeted
I think parsers are uniquely good for this kind of thing since they (often, not always) have clear specifications and available conformance suites, plus they can be heavily fuzzed and differentially tested. But like, should we just do this for everything? Why or why not?
6
2
2
59
4,383
Patrick Smith retweeted
hello, Finder 1:1 in GPUI using native components and liquid glass
45
19
6
638
55,171
Patrick Smith retweeted
As of this month, The Consensus has now passed 1,000 followers on Twitter, 1,000 followers on LinkedIn, and 1,000 newsletter subscribers. A decent first 9 months! Even in the age of AI, and likely supercharged due to it, people still want to understand how software works.
9
2
1
109
5,290
Patrick Smith retweeted
What’s the evidence on rent control? I charted the findings from 112 studies: Great for rent-controlled tenants ✅ But: Pushes up rents for other people ❌ Pushes down new housing construction ❌ Pushes down building quality ❌
79
732
116
2,954
259,688