Competition !
Meta just came out with a competitive coding agent, Muse Wonder Full. You can subscribe for either 50$/month or 200$/month. For coding, 'Muse Wonder Full' seems to be a match for the best models from Anthropic and others.
Meta just came out with a competitive coding agent, Muse Wonder Full. You can subscribe for either 50$/month or 200$/month. For coding, 'Muse Wonder Full' seems to be a match for the best models from Anthropic and others.
10
0
31
2,972
Do you trust Claude Code or Grok to run on your machine? How do you know it won't email all your secrets to your girlfriend?
Microsoft cam up with a solution: Microsoft Execution Containers (MXC) . The idea is simple. You write a policy, a JSON file that says which files the files it may write, and whether it may use the network.
It is not a bad idea.
However, on my macBook, I find that MXC can add significant overhead. I measure over 10 ms in startup cost.
It gets worse if you run a command like git which seeks to access a temporary directory that you did not allow. Then the whole thing just stall.
So for many commands, you probably want to set the TMPDIR variable to a location where your MXC can write. Even so, there is still a visible penalty.
It may not matter in the grand scheme of things if you can avoid the worst slowdowns: what are a few milliseconds? In my experience, it does add up and lower your quality of life. I don't know if it will the case with MXC.
Microsoft cam up with a solution: Microsoft Execution Containers (MXC) . The idea is simple. You write a policy, a JSON file that says which files the files it may write, and whether it may use the network.
It is not a bad idea.
However, on my macBook, I find that MXC can add significant overhead. I measure over 10 ms in startup cost.
It gets worse if you run a command like git which seeks to access a temporary directory that you did not allow. Then the whole thing just stall.
So for many commands, you probably want to set the TMPDIR variable to a location where your MXC can write. Even so, there is still a visible penalty.
It may not matter in the grand scheme of things if you can avoid the worst slowdowns: what are a few milliseconds? In my experience, it does add up and lower your quality of life. I don't know if it will the case with MXC.
24
2
66
6,904
Thanks to everyone who gave me ideas as what to talk about regarding AI and programming in my upcoming lecture. I will try to give credit.
I am giving a talk on AI and programming in a couple of weeks. I have an abstract but the talk is not ready yet.
What should I talk about?
The audience will be made of young people.
1
1
26
2,161
I am giving a talk on AI and programming in a couple of weeks. I have an abstract but the talk is not ready yet.
What should I talk about?
The audience will be made of young people.
What should I talk about?
The audience will be made of young people.
25
4
58
6,115
Back in 2023, I compared the three main JavaScript runtimes (Node, Bun, Deno) on a WebSocket benchmark.
I have redone the benchmark with a better methodology (had my AI write a client in C).
Surprisingly to me, I find that Node.js really does well. Deno : not so much.
Details in the article.
I have redone the benchmark with a better methodology (had my AI write a client in C).
Surprisingly to me, I find that Node.js really does well. Deno : not so much.
Details in the article.
13
2
93
12,392
This is a representation of what an orbital data center might look like. It stays close to the surface of the Earth, and it can always face the Sun.
2
0
10
2,102
If you are an engineer who builds actual software and you'd like to review a research paper from time to time, please sent me a note in X chat. You *DO NOT* need to have a PhD. But, ideally, you should sometimes enjoy reading a research paper. Give me a description of your tor for Software: Practice and Experience. I filter papers that I send to review, but I often run out of people with actual experience. I have a few names, but they get exhausted after a time.
Yes, I know I could just ask Claude to review the papers... and it does a fair job... But what I care about is whether human beings find the paper useful.
(This is free labor, something you have to be willing to do for fun.)
(The system may yet collapse in the near future as AI can write and review conventional papers much better than we can. For the time being, I am enjoying the ride and trying to keep things meaningful and rigourous.)
Yes, I know I could just ask Claude to review the papers... and it does a fair job... But what I care about is whether human beings find the paper useful.
(This is free labor, something you have to be willing to do for fun.)
(The system may yet collapse in the near future as AI can write and review conventional papers much better than we can. For the time being, I am enjoying the ride and trying to keep things meaningful and rigourous.)
4
6
51
2,760
There is much talk about 'AI cloning'. You take an existing piece of software, and you build your own version... using AI as a helper.
It is amusing, but it is akin to using the newly invented radio to broadcast theatre plays.
I am not sure that kids in the future will want to ll need software forever... but do we need *this* particular kind of software?
It seems almost certain to me that what will eat into Adobe's margins won't be a AI clone of their software.
And to be clear, I think that such clones will, eventually, work quite well. Don't be too quick to judge them on their current failures. They will quickly become better if people care.
Yet the future is not some open source version of Microsoft Teams or Adobe Photoshop.
We now have the power to invent entirely new user experiences... millions of people can run experiments, can embed photo editing into their AIs, just the way they like it...
AI cloning is the obvious and boring part.
It will get wilder. Far wilder.
It is amusing, but it is akin to using the newly invented radio to broadcast theatre plays.
I am not sure that kids in the future will want to ll need software forever... but do we need *this* particular kind of software?
It seems almost certain to me that what will eat into Adobe's margins won't be a AI clone of their software.
And to be clear, I think that such clones will, eventually, work quite well. Don't be too quick to judge them on their current failures. They will quickly become better if people care.
Yet the future is not some open source version of Microsoft Teams or Adobe Photoshop.
We now have the power to invent entirely new user experiences... millions of people can run experiments, can embed photo editing into their AIs, just the way they like it...
AI cloning is the obvious and boring part.
It will get wilder. Far wilder.
11
4
89
6,059
Find bugs in our fast URL parser and get rewarded!
This URL parser is used everywhere in critical code.
C++
This URL parser is used everywhere in critical code.
C++
📢📢📢 hey chat!
@yagiznizipli and I decided to announce a bounty for vulnerabilities in Ada-URL.
It's a url parser project written in c++ by my man @yagiznizipli
We'll pay up to $1k.
github.com/ada-url/ada
4
0
62
3,681
I write a lot about data parallelism. About how our processors have special instructions that can do several operations at once. They are often critical to get the best performance out of a CPU.
Compilers are sometimes able to take advantage of these instructions. But it is hard rk.
For Java programmers, you can write directly for data parallelism since JDK 16. With enough skill, you might multiply the performance of a few key functions.
The Vector API is in the incubator module jdk.incubator.vector. It still not officially supported, but you can enable it by passing --add-modules=jdk.incubator.vector. I expect that it will soon become a mainstream feature in Java.
Roman Snytsar wrote a book about it: Mastering SIMD with Java Vector API, and the subtitle is Unlocking Single-Core Performance Through SIMD Optimization. I had the honor of being the technical reviewer. I learned a few things while reading the book and I enjoyed it.
Who has time for books, especially technical books?
What a book offers is time to reflect. Reading technical books today is probably just as relevant as it ever was. The author takes you on a story... and gets you to think.
Snytsar's book works from problems like sum an array, compute a mean and a standard deviation, remove duplicates, merge two sorted arrays, and so forth. A lot of them are similar to the problems you will encounter as a programmer. He starts from the ordinary solution and rebuilds it with the Java Vector API. He then runs benchmarks. He looks at the assembly code.
The book covers performance issues such as unrolling, asymmetric loops, structural hazards, dependencies. Snytsar shows us that, in some instances, data parallelism can fail to bear fruits. The negative lessons are just as important as the positive ones.
The Java Vector API is a thick layer of abstractions, but Snytsar shows that some tricks work better on some hardware than others. I liked the chapter on removing duplicates, which is built on compress. It is one instance where the specific hardware matters. And you are unlikely to just find out about it yourself.
It is a book worth buying if you are a Java programmer with a focus on performance.
The book: Roman Snytsar, Mastering SIMD with Java Vector API, Apress, 2026.
amazon.com/dp/B0GPBQNTJ5
(I do not get a cent if you buy the book.)
Compilers are sometimes able to take advantage of these instructions. But it is hard rk.
For Java programmers, you can write directly for data parallelism since JDK 16. With enough skill, you might multiply the performance of a few key functions.
The Vector API is in the incubator module jdk.incubator.vector. It still not officially supported, but you can enable it by passing --add-modules=jdk.incubator.vector. I expect that it will soon become a mainstream feature in Java.
Roman Snytsar wrote a book about it: Mastering SIMD with Java Vector API, and the subtitle is Unlocking Single-Core Performance Through SIMD Optimization. I had the honor of being the technical reviewer. I learned a few things while reading the book and I enjoyed it.
Who has time for books, especially technical books?
What a book offers is time to reflect. Reading technical books today is probably just as relevant as it ever was. The author takes you on a story... and gets you to think.
Snytsar's book works from problems like sum an array, compute a mean and a standard deviation, remove duplicates, merge two sorted arrays, and so forth. A lot of them are similar to the problems you will encounter as a programmer. He starts from the ordinary solution and rebuilds it with the Java Vector API. He then runs benchmarks. He looks at the assembly code.
The book covers performance issues such as unrolling, asymmetric loops, structural hazards, dependencies. Snytsar shows us that, in some instances, data parallelism can fail to bear fruits. The negative lessons are just as important as the positive ones.
The Java Vector API is a thick layer of abstractions, but Snytsar shows that some tricks work better on some hardware than others. I liked the chapter on removing duplicates, which is built on compress. It is one instance where the specific hardware matters. And you are unlikely to just find out about it yourself.
It is a book worth buying if you are a Java programmer with a focus on performance.
The book: Roman Snytsar, Mastering SIMD with Java Vector API, Apress, 2026.
amazon.com/dp/B0GPBQNTJ5
(I do not get a cent if you buy the book.)
3
12
114
5,512