I'd love to hear more about your setup. The last time I tried messing with virtual webcams on macOS, most of the OSS options were still using the deprecated DAL API, which stopped working in Sonoma (without setting a `system-override` from Recovery Mode).

Do you know if your setup is using the new CMIO APIs? Or did you need to set the `legacy-camera-plugins-without-sw-camera-indication` override?

I believe the old DAL stuff is gone, maybe somewhere around version 30? I used to have awful crashes all the time, either Zoom's capture host process (caphost) would crash, necessitating completely restarting the app, or OBS would crash. Now it just installs a camera extension, so yes, I believe it's all on the CMIO APIs.

I am mostly broadcasting the desktop view of a second or third computer on my desk, plus slide decks, graphics, etc. I hate Zoom's clunky screen picker UI and never want to accidentally share plaintext secrets or PII to the attendees by choosing the wrong window or absentmindedly moving a window to a shared monitor or something, so I got around that by just using OBS as a switchboard, and never having production data on my test computers. One is hooked up to an Elgato Neo capture device over HDMI, the other when required over a BMD Intensity 4K on an eGPU dock. Slide decks are the only data from my presenter computer ever shared, but via a 'macOS Screen Capture' source. I have a few scenes with text only too for info/time stuff. It's great being able to slap some text over a scene, or have two computers up at once in a half-and-half arrangement. Plus plugin support (used to capture from the BMD from a PC tower and send the video to my presenter computer over NDI when an older elgarbo capture card shit the bed), audio and video filters/effects, goddamn shaders, sky's the limit since OBS is so extensible. I've done in person presenting and still used OBS to generate a fullscreen projector of the program output.

I also personally have always found the "presenter awkwardly clicking around through an un-played Powerpoint" to be really unprofessional. I can hide all of that messy stuff and not just a bunch of dead air.

Right now I'm working on getting an Adafruit Macropad to talk to an ESP32 over UART so I can build a wireless streamdeck that doesn't require any shitty bloated control software. The ESP32 would then send commands/get statuses from the Websockets server and send em back to the Macropad so it can update the screen. This would be impossible to do with Zoom since the app isn't scriptable (on macOS most AppKit UI can be automated with AppleScript) and they only engineer for the lowest-common-denominator use cases.

Bill is typically not a given name at all, but rather a nickname for someone named William (William -> Will -> Bill).

You might as well complain about Betamax. XHTML is not the future.

Wait until you learn what an iOS app's .ipa file is.

What’s that saying, there are 3 kinds of files?

zips, text files, and binary files

It's not a "feature" for the 89% of users whose SSD capacity was being wasted.

"Prefer `xor a` instead of `ld a, 0`" is basically the first optimization that you learn when doing SM83 assembly.

https://github.com/pret/pokecrystal/wiki/Optimizing-assembly...

That's correct. I ran a special metadata VDEV 3-way mirror using this NVMe PLX card for a while https://imgur.com/a/xiwzkA6

Google, through YouTube and YouTube TV, already runs one of the most significant video processing lines of business in the world. If they had any interest in supplanting FFmpeg with their own software stack, they wouldn't need to muck around with CVEs to do so.

Not for themselves, but for everyone else for some reason.

Though it's more likely there's just some team tasked with shoving AI into vulnerability hunting.

Before Epic Games acquired it...

Docling primarily uses AI models to extract PDF content, this project looks like it uses a custom parser written in Java, built atop veraPDF.

Correct me if I am wrong, but Docling can do both. It has also, among other strategies, a non-AI pipeline to determine the layout (based on qpdf I believe). So these projects are not that different.

While it has a PDF parser, my understanding is that it is mainly used to break a PDF document into chunks, which are then handed off to various specialized models. From its docs: "The main purpose of Docling is to run local models which are not sharing any user data with remote services."