sly010

Deterministic AI.

Aka, a computer program.

Beware, there are versions of go where sendfile is broken and only sends the first 4k of a file on macos.

people over-rate how much software/IP is useful in running a successful business

Indeed, by a couple trillions...

I don't think people doubt what AI _could_ do, they just have been through enough enshitification cycles to know this is not any different. Right now AI is better than Google but only because Google regressed so much. Market forces always prevail. The operating costs are just too high to offer AI for free for everyone but people will refuse to pay, so AI (at least for the masses) will become just an other marketing funnel companies can buy out. I also don't see how AI will change the fact that clothing companies target average users and don't serve the long tail.

No batteries!? Go has a huge stable standard library no other language even comes close to. Built in tooling for unit testing, performance testing, debugging, code formatting, package management, etc. And most go binaries can be compiled statically so libc is not even a dependency. Golang is the definition of batteries included.

Then of course there is the whole fractal nature of software. As you add more detail, the arrows can flip flop around. Polling in a lower level of a stack can very much be used as a push mechanism. (e.g. USB interrupts are in fact the host polling the device)

This is why communicating architecture is often as hard if not harder than implementing it.

There is in fact a 3rd: build time dependencies.

Control: Object/module/function A calls object/module/function B.

Data: The call can either push or pull data.

Build dependency: The call can be direct (A depends on B) or indirect through an interface/callback/etc (both A and B depends on the interface).

Ideally every design document includes all 3 as separate diagrams.

... or the WM loads the compositor, or the WM links to a compositor library (i.e. wlroots). The point is there are options...

Honestly, every time this topic comes up, I feel like the person complaining just doesn't want to put in the work and they are angry that they don't get an easy win. And maybe that's a good thing. Do we really need more half baked WMs?

chosen to use fear and intimidation to help sell the agenda of the big tech CEOs who, in turn, have somehow managed to use coal-fired GPUs to capture society’s output and sell it back to us, while converting a significant portion of the economy into an expanding envelope of hot gas

I work for a very big tech company and I have no one to share this with.

There is a 3rd party Android app that uses the accessibility APIs to (supposedly) track and limit my short video use. However, it's broken, so I can't watch short videos at all :)

We get it. You are a manager.

I've seen this image generated by meta AI. The prompt was something like: think of a room, make it look like anything you like, but do not in any circumstance put a clown in it. Guess what...

I think Jason has a "do not think of an elephant" problem.

Sorry for the snark, but we couldn't even do this for humans, but let's do it for poor poor LLMs? It's kind of ironic that NOW is the time we worry about usability. What happened to RTFM?

Genuine question: How does this work? How does an LLM do object detection? Or more generally, how does an LLM do anything that is not text? I always thought tasks like this are usually just handed to an other (i.e. vision) model, but the post talks about it as if it's the _same_ model doing both text generation and vision. It doesn't make sense to me why would Gemini 2 and 2.5 would have different vision capabilities, shouldn't they both have access to the same, purpose trained state of the art vision model?

Nah. You misunderstood. "They" don't make money on human time wasted. They make money on ads served. They don't particularly care if the ads were served to humans or agents, they get paid either way. Bot-traffic is actually good for tech companies because it inflates numbers. Capthas are not there to waste our time, but are there to improve their credibility ("We are certain those ad-clicks were real humans because the captha said so").

First off, I always thought the type of things described (tracking mouse movements, keypress jitter, etc) are already done by ReCacpha to decide when to present the user with a captcha. I am surprised they are not already doing this.

Second, I am surprised AI agents are this naive. I thought they would emulate human behavior better.

In fact, just based on this article, very little effort has been put into this race on either side.

So I wonder if is has to do with the fact that if companies like google reliably filtered out bot traffic, they would loose 90% of their AD revenue. This way they have plausible deniability.

Math has a PR problem. The weight being non-uniform makes this a little unsurprising to a non-mathematician, it's a bit like a wire "sphere" with a weight attached on one side, but a low poly version. Giving it a "skin" would make this look more impressive.

And just like optimizing compilers LLMs also emit code that is difficult to verify and no-one really understands, so when the shit hits the fan you have no idea what's going on.

Very cool, but by css-rotating (skewY(-6deg)) the canvas at the last moment, you introduced aliasing on the border between the canvas and the rest of the page which kills the vibe. The browser can't automatically blend the canvas with the rest of the page. It's noticeable even on a brand new retina display. Maybe you could keep your canvas square and introduce the skew in the shader.

Isn't that just true for vulnerabilities in general? Trust Zone is not a security mechanism, it's an isolation mechanism.

I don't disagree, but funny that I recently made a point to someone that modern consumer embedded systems (with multiple MCUs connected with buses and sometimes shared memory) are basically small distributed systems, because partial restarts are common and the start/restart order of the MCUs is not very well defined. At least in the space I am working in. (Needless to say we use C, not rust)

The goal of programming is not to write code (however much I enjoy that part), it is to solve problems.

Right, which is exactly why autocomplete is not a huge help.

FWIW ink is not expensive, but traditionally the nozzles and electronics and the printhead are all bundled with the cartridge, making it expensive. The big companies now offer inkjet printers with ink and print head separate (eg Canon pixma) and for those printers the ink (in a literal bottle) is priced much more reasonably.

There are other issues of course, I am not trying to give them slack here. If you don't use self cleaning once a week, the printer head clogs and you have to replace it anyway. If you use self cleaning too much, the waste ink reservoir fills up and your whole printer is toast unless you reset the firmware somehow. But there is a whole community for those kind of things.

Or you could say let's not normalize broken AI in health care related systems, but potato potato I guess.

I'm sure somebody has written a book already about how ostensibly wealthy societies can fail at basic infrastructure that they previously mastered, driven by greed and complacency and other socioeconomic factors.

How about Foundation by Isaac Asimov.

Hah. At a previous place I found that our cloud cost consisted of 90% storage costs. The data? Tens of thousands of incomplete backups of the in-office file server. 3 years of the NAS continuously trying to back itself up to S3 and failing every time.

I feel you. Protobuf is the organizational equivalent of adding more lanes to congested roads or mandating more parking spaces. It leads to more cars.

Sounds about right. Hiring pipeline is probably too long to guarantee anything, but most people don't work out, so recruiters want to keep good leads in the pipeline as long as possible, just in case something comes up. Never attribute to malice that you can attribute to (in this case organizational) incompetence.

Note that I meant "handling (forward) pressure (or preventing back-pressure) can be done by simply having your system be performant enough. Of course this is not a dynamic property you can adjust at runtime, just wanted to add for the sake of completeness, because it should be part of the mindset. Sometimes a database falling over simply needs a few queries optimized.

You could have back-pressure in the most optimized system.

Most batch processing systems. But you don't want back pressure in interactive or real-time systems, like graphics, gaming, audio, cars, planes, or even just real time collaboration systems (e.g. Figma). In all these cases back pressure is to be avoided.

Edit: I think the article also misses one of the most important ways to release pressure. And that is scaling throughput either horizontally (e.g. by adding more servers) or vertically (e.g. by optimization of software)