There just wasn’t really user-submitted content back then. If you wanted to post something on the internet, you had to create a website.

There were tools that made a website a bit easier to create (most notably Geocities), but it was still a fairly technical task to simply get text that you typed show up on a website accessible to all.

Simply being able to reply to someone else? That wasn’t really on the web, although by 2001 there were a few attempts to bring that kind of forum functionality to the broader web on actual http/www webpages (when it was previously on Usenet and BBSes and mailing lists and the technologies known as the internet before and separate from the world wide web). vBulletin was the popular one, and I think phpBB was fairly popular, too, but those projects started in 2000 and weren’t that popular in the first years. Web hosts actually offering a full blown LAMP stack (Linux, Apache, MySQL, PHP) for hosting, and the technical skills to administer that kind of service, even on free software, was a pretty big hurdle to adoption before 2003 or so.

In 2001, almost all of the web was static content, with no method of interacting with the content other than simply viewing it. You’d click on links to retrieve another piece of content. By 2005 the web 2.0 revolution was underway, and websites had become interactive, with dynamic content drawing from relational databases under the hood. But in 2001? No, the web was pretty static.

Your description of the rise of the social algorithms is part of the story since 2010, but even that leaves out the period of time in between.

HTML / JS is still a buggy pile of slop.

In what world is JavaScript considered buggier than Flash, either in the actual content in the wild or the client-side software rendering/running that content? It wasn’t true when Flash died and certainly isn’t true today.

Flash was a security nightmare, with all sorts of kludges patched on to try to deal with fundamental flaws in how it handled privileges. If you want to go back to the days where zero click exploits can take over your machine just from a browser visiting the wrong URL, leave the rest of us out of that vision.

The current web experience is complete garbage in comparison.

Mm, and what makes you think that handing Adobe the keys to control everyone’s web experience would make it better?

As a browser-supported web image format, the only real runaway advantage of JXL is the ability to losslessly reencode existing JPEG files, when the vast majority of image files that people already have are stored as JPEG “originals.”

But as an overall image format, JXL has a much more ambitious scope: much higher limits in the spec to resolution, bit depth, layers, etc., showing an intent to be used as a raw image capture format and printing format, not limited to screen resolutions and bit depth like AVIF is.

So if the original file gets stored as JXL, the workflow and pipeline of a JXL native process the whole way may have an advantage over exporting to a screen-friendly AVIF at the end of the process, while the original still gets stored as another format.

Re-encoding from JPEG to AVIF will always drop quality. Lossless re-encoding from JPEG to JPEG XL makes a pixel perfect reproduction, while shrinking file size.

Apple’s Safari supports still images in JXL.

Perhaps more notably, Apple’s iPhone ProRaw supports JXL encoding for its raw sensor data. They’re laying the groundwork for JXL to be a supported format from the point of capture at the camera, through all the processing to make an image for web publishing or whatever.

The killer feature is that JXL is better in that it can losslessly encode JPEG further, and the overwhelming majority of the legacy image files that people have are JPEG. That alone should justify its support, because there are a lot of files out in the world where the highest quality, closest to “original” quality file is stored in JPEG format. A format that allows for the further compression with zero loss of quality from those originals is really important.

And the other thing this article (and a lot of the discussion around JXL) chooses not to cover is how JXL is a good format outside of just web images. It’s not just looking to replace JPG/PNG/webp. It’s also looking to replace raw photography formats like DNG, TIFF, and other formats that are used for full workflows from image capture from the imaging sensor itself, from cameras to scanners to medical imaging.

If JXL succeeds at becoming the dominant raw capture format, the entire workflow of processing those raw images into exported web-friendly images will favor JXL for photography.

No, you’ve got it all wrong. Flash was buggy and insecure proprietary software written for proprietary OSes. I’m glad it died, after several years of trying to fight its normalization on the web, where you’d need to install shitty plugins like gnash on Linux to try to see the same web that people were seeing on Mac or Windows.

The web should always be built on open standards, and Flash was the biggest barrier to that ideal at the time that iPhones came out. Flash deserved to die, and I celebrated when it did.

Hmm I don’t think you would be a good choice to be appointed as the champion standing up for human emotion.

Steps that either would not work between architectures or that Rosetta would handle (thereby answering your own question)

What? They would compile the code that they control for the M-series architectures. The target architecture is M-series ARM chips running on MacOS.

Rosetta is for translating x86 to ARM. Which wouldn’t be necessary because there are no x86 binaries involved at all.

I don’t understand why Rosetta has to be involved in any way at all.

Apple provides APIs for programming low level GPU instructions, including AI workloads, on their chips, using the Metal API. Anyone interested in using Apple hardware to its full potential can just write Mac-native software, same as any other MacOS native software directly compiled as binaries to run on Mac M series chips. No Rosetta required for translating x86 instructions to ARM, when the whole thing compiles for Apple’s native instructions in the first place.

In 2011, Motorola split into two companies, each with rights to the Motorola name and logo.

Motorola Mobility took the business lines related to cell phones. It got bought by Google, and then Lenovo.

Motorola Solutions took the business lines related to everything else, including (and especially) radios for law enforcement, fire departments, militaries, and surveillance video tech.

Yup. My main living data is on a NAS, my on-site backup is an external USB enclosure with a SATA hard drive hooked up to a Mac Mini, all so that I can leverage a Backblaze account to back up the external hard drive with unlimited data.

Wait can tokens just be directly compared like that? My impression is that token cost can vary by 2 orders of magnitude, depending on model, because the actual work of computation varies by that much between models.

She also did this while she was still a full time undergraduate university student. Normally you’d expect someone with more experience to be doing this kind of low level hardware hacking, but she did all this while still young, before even graduating from university.

Reportedly, the app uses an on-device machine learning model to classify the content.

Using your shiny new phone to train AI…

I would think the models are already trained, and then deployed to phones to do on-device inference. The models should be fixed and static by the time they end up on a phone (in large part because phone hardware is just not technically suitable for training, whether pre training or post training).

True, although that’s kinda on the parents for naming their kid after a character whose arc isn’t finished in popular culture.

Imagine being named Alexa, born in 1998, minding your own damn business when Amazon just completely wrecks your name as you’re entering high school.

Yes, the rules of Conway’s Game of Life is Turing Complete.

The glider shape (the thing moving diagonally out of the window in this meme) can be created non-destructively by a glider gun (another shape that creates gliders), which can be controlled using stored pixels in another configuration within the glider gun, including taking in input from other gliders, such that the whole thing can loop and do everything necessary to be a universal Turing machine. Not super efficiently, but it will still do the things that Turing machines can do, given enough cycles.

The -cel suffix suggests that it’s an insult (derived from incel).

Vibe coding is the name for creating computer code by telling an AI to generate the code for you, without necessarily even understanding anything about the code itself.

A thousand yard stare is a description for someone who looks like they’re staring off into the distance, as if they’re having some kind of PTSD flashback, or just at a loss for words.

This is a photograph of Sam Altman, the CEO of OpenAI, a major generative AI company.

Tokens are the unit of measure of how much computing power a particular generative AI query uses.

So basically, it’s a meme that ironically takes the position of the vibe coder, mocking someone who is actually unwilling to use the AI code generation tools (calling them a codecel and criticizing them for saying something anti-vibe-coding or vibephobic), by giving the blank stare of Sam Altman’s soulless eyes, and calling it a billion token stare, but basically doing it ironically to make fun of generative AI enthusiasts by lobbing a really stupid criticism and making the critic look bad.

The mechanism happens in the way LLMs generate text sequentially. So as the LLM generates text, each time the LLM encounters a choice between two or more equally probable words and has to pick one, it chooses from a known pseudorandom key instead of its previous practice of actually relying on a source of randomness.

If you run the text through a checker script that tries to predict the next word, and takes note of every choice made in any given fork in the probabilistic flowchart, you can see that whole sequence of choices and compare to whether it matches a known key.

You’d need to know the model itself well enough to know when the text reaches a particular fork, and what the choices are in that fork, and the key itself, so even OpenAI/Google/Anthropic may not be able to run this watermark detection on their competitors’ models. But they’d be able to know it themselves, for their own models.