slimsag

Unless the best models themselves are costly/hard to produce, and there is not a company providing them to people free of charge AND for commercial use.

Congrats Mitchell! It has been really cool to see Ghostty progress as a project, and I've enjoyed having it as my daily driver these past few years :)

While there is criticism of MBTI, it can actually be quite useful in sparking interesting conversations about work styles and preferences. It also helps people reflect on their own tendencies/behaviors in general.

I won't pretend its perfect, but it can be a fun team-building exercise and provide common language for discussing differences.

We are working on a stealth startup which will use AI to analyze e.g. Slack messages and build out results of personality tests using methods like MBTI, so that e.g. when reaching out to someone in your workplace, you can get view a personalized overview of how to interact with that person, their behavior, how they might prefer you communicate with them, etc. Email in my bio for anyone interested, we're looking for early adopters.

SDL provides..

..an image decoding library.

..an audio input/output library.

..a multi-channel audio mixer.

..ttf and rtf font rendering.

..networking.

..runtime shader cross-compilation

..its own entire graphics abstraction layer / API.

..its own shader language, shader compiler, and shader bytecode format.

I don't know at which point something becomes 'high level', but SDL is only 'minimal' if you use a subset of its functionality.

I am referring primarily to the fact that a restricted subset of WebGPU is needed ('compatibility mode') to support D3D11 / GLES era hardware[0]

[0] https://github.com/gpuweb/gpuweb/issues/4266

This is the right answer^

But even more, the level at which WebGPU exists (not too high level, not too low level) necessitates that if a native API graphics abstraction sticks with the WebGPU's API design and only 'extends' it, you actually end up with three totally different ways to use the API:

* The one with your native 'extensions' -> your app will only run natively and never in the browser unless you implement two different WebGPU rendering backends. Also won't run on Chromebook-type devices that only have GLES-era hardware.

* The WebGPU browser API -> your app will run in the browser, but not on GLES-era hardware. Perish in the verbosity of not having bindless support.

* The new 'compatability' mode in WebGPU -> your app runs everywhere, but perish in the verbosity of not having bindless, suffer without reversed-z buffers because the underlying API doesn't support it.

And if you want your app to run in all three as best as possible, you need to write three different webgpu backends for your app, effectively, as if they are different APIs and shading languages.

Unfortunately not everyone shares that opinion of their restricted use-cases.

I've seen ~100 line HTTP handler methods that are implemented using generics and then a bunch of type-specific parameters inevitably get added when the codepaths start to diverge and now you've got a giant spaghetti ball of generics to untangle, for what was originally just trying to deduplicate a few hundred lines of code.

The article seems to focus on YubiKeys (even despite using a mac device), but at least in the Apple ecosystem the state of the art is far ahead of plugging in a thumb drive style device IMO.

When I navigate to various websites that support Passkeys these days - Lowes, Home Depot, and handfuls of others - on iPhone (firefox) I am prompted with 'Do you want to use Face ID to sign in?'

If I do the same on macOS (firefox or safari, I'm sure Chrome does it too.), I am similarly prompted 'Do you want to sign in using a passkey?' and it gives me steps to use my mac's fingerprint reader or delegate to the iPhone Face ID to sign in on my Mac.

Combine that with Apple offering to use a private/hidden email address for signing up to the service in the first place and forward mail to your real email, plus the auto-generated secure passwords stored in the new Apple password manager.. and it's a pretty magical and secure experience as a user.

...and once you have a threads account, the UI in instagram pushes you to one-click accept cross-posting everything to Threads

Reminds me of a Zig proposal I saw recently to make the std.Thread.Pool API respect the Linux jobserver and macOS dispatcher automatically out of the box:

https://github.com/ziglang/zig/issues/20274

Javascript in browsers already has a full atomics API:

https://developer.mozilla.org/en-US/docs/Web/JavaScript/Refe...

I'm not sure why Web Locks is useful TBH. I guess if you don't understand atomics it's a friendlier API?

The real solution is games designed for playing with friends and treat all non-friend players as potentially malicious.

Early first-person shooter games had this figured out (small servers with 20-30 regular players, the server admin could choose to ban you), RTS games have this figured out, many MMOs have this figured out (interact with non-friends sometimes, but they have to 'join your party', etc.)

Playing with random strangers on the internet who may want to grief/destroy your game, be incredibly toxic, or cheat against you in general.. that's the cost of playing with random people in a completely public forum.

Lots of possibilities here.

People are learning to prompt LLMs in ways that produce better results for their LLM of choice, so switching to another one they find their approach no longer works as well.

Or.. LLMs have different personalities in terms of output; some being more or less direct/polite than others, or sounding more or less confident; and that is causing people to perceive a difference that in terms of factual answers may not be different.

Or just personal preference masquerading as intelligence - a classic among software engineers.

Has anyone worked on making tokens 'clusters of words with specific semantic meaning'?

e.g. instead of tokens ['i', 'am', 'beautiful'] having tokens ['I am', 'beautiful'] on the premise that 'I am' is a common set of bytes for a semantic token that identifies a 'property of self'?

Or taking that further and having much larger tokens based on statistical analysis of common phrases of ~5 words or such?

Is there something similar for GCP?

it uses absolute difference epsilon equality ('close enough to be considered equal'):

    static int pretty_float_equal (float a, float b) { return fabsf(a - b) < FLT_EPSILON; }
    static int pretty_double_equal (double a, double b) { return fabs(a - b) < DBL_EPSILON; }
    static int pretty_long_double_equal (long double a, long double b) { return fabsl(a - b) < LDBL_EPSILON; }

In 2021 Automattic, the company that effectively owns Wordpress, has a valuation of $7.5 billion in 2021, and revenue of $750M in 2024. It's big money.

MoltenVK doesn't support geometry shaders for ~3 years now[0]

[0] https://github.com/KhronosGroup/MoltenVK/issues/1524

Interestingly, public U.S. state property records will just disclose where you live whether you like it or not. With as little as your name, a home address is trivial to find.

The attorney general’s office consumer protection division in your state does what most people think the BBB does.

BBB is just a review website. Like Yelp.

Filing a complaint with BBB is like saying 'I left a bad Yelp review' ... useful, maybe.. if the company cares..

I haven't been able to get llama.cpp's mmap logic to work on macOS

Its close in spirit to DRM: they enforce that if you are using their projects, it is under their legal terms.

Many of their VS Code extensions[0] have license terms which /prohibit/ them from being used in editors that are not Microsoft's VS Code editor, and Microsoft can issue cease-and-desist for violations of this, as well as 'open source forks of VS Code' using their marketplace services.

Microsoft also has a pattern they follow where they publish repositories on GitHub to proclaim they are open source, while actually distributing proprietary binary blobs in them that are fundamental core parts of functionality to the codebase[2][3][4]

[0] https://github.com/microsoft/vscode-cpptools/issues/6388#iss...

[1] https://github.com/microsoft/vscode-cpptools/issues/5784

[2] https://github.com/microsoft/win32metadata/issues/766

[3] https://devlog.hexops.com/2024/building-the-directx-shader-c...

[4] https://github.com/microsoft/Azure-Kinect-Sensor-SDK/issues/...

This is better indeed, but from recollection last time I tried this.. I think that still requires a full OpenJDK installation and I believe those `sdkmanager` commands install multiple GB of tools sadly.

Selling to businesses is an easier way to make large profits in general, definitely. A single sale can be worth millions, and you have fewer stakeholders that you have to convince - especially if you already have contacts in those businesses or can get introductions easily.

Selling to consumers means each one will judge and pay only for the value they see from your product; which means everything hinges on your product's value to many individual users, rather than a small group of stakeholders looking for a very specific set of benefits (ones you can often build for as they ask for it!)

One major reason B2C companies charge a subscription fee is because they need to show to investors that the product provides customer lifetime value: not a one-off value-add, and that there is a grow-able user base locked in to the product.

You cannot compete with free, unless their experience is so bad that people are willing to throw money at the problem to have it be better. Most of the time, that isn't the case. 'Facebook that values your privacy' is not a value proposition that most people care to pay money for.

'Photoshop but [cheaper, free-with-ads, without an Adobe subscription]' is an area where people have successfully built businesses, however.

We are moving past the “Airbnb for dogs” era and into the “build a better quantum computer” era.

I disagree. Most software still sucks for the average user.

Every app comes with its own nickel-and-dime subscription fee. You want to make a photo collage of your dog? That's $10/mo for PHOTO COLLAGE PRO, thanks.

You want to have a nice team-wide issue tracker / project planner? That's $12/user/mo, and contact our sales team if you need data security.

Using our free tier? No problem, please enjoy as we use our /alternative means/ to monetize you. Feel free to watch this ad while you chew on whatever that means.

..all this to say, there is still very much an opportunity for a software race-to-the-bottom, where people provide equal or better value to users, for equal or less monies. Pick any business, and you can do this too, as long as it is not a market where the leader is operating at a loss for dominance and as long as your execution is good enough.

DJI sells a product called Aeroscope which is an antenna platform used by airport towers; it shows them via the frequencies the drones operate on where the drone is, that it is active, all information tied to the serial number including the drone pilot's FAA registration (who the owner of the drone is, the pilot's phone number, their address, etc.) It also this shows them the location of the pilot via the remote control frequency.

This applies even for their sub-$1000 consumer drones, and FAA registration is mandatory for all drones above 250g weight.

In FPV communities, DJI is considered the closed/proprietary ecosystem that locks you in. Even their product lines are not exactly compatible with each-other.

You can pick up Walksnail-protocol or HDZero-protocol goggles for ~$500 new, and HD 4k drones compatible with them like Caddx Gofilm 20 for $400 - and connect any controller which uses an open-source ELRS protocol.

Granted, all of the hardware is manufactured in China I'm sure - but the software is miles apart.

Nostr is a protocol. Bluesky is a product.

The other reasonable course of action would be to go to court, where a jury of average citizens could determine whether or not Apple or Docker is in the wrong and decide damages.

Docker wasn't pushed into taking the image down, they *decided* that they'd rather take the image down than have a jury decide who is in the wrong.