But it isn’t encoding knowledge, it’s encoding word correlations.

I’m saying that humans do this a lot, too. Qualitatively, it’s different, in that this particular batch of frontier LLMs will get things wrong in ways that most human brains wouldn’t, but as a category of error it’s not unique to LLMs.

I know a ton of facts that I learned only through reading, and have no actual firsthand knowledge/experience or ability to test it: Jupiter is larger than Saturn, the atmosphere during the Carboniferous period was high in oxygen, cigarettes cause cancer, Thomas Jefferson owned slaves, the capital of Norway is Oslo. At best, I can cross reference other sources and see that things are consistent with each other. Is my belief in those facts “knowledge,” or is it merely recognizing from my training data that those particular words can validly be presented in that order?

If you ask average people on the street whether FAT32 is a good filesystem for a 64GB removable drive, most of them won’t know, but there are a handful of bullshitters who might confidently parrot back things they can Google but not understand. That’s part of the human condition, too.

I’m by no means an AI booster/enthusiast. I suspect LLMs/transformers are actually a dead end, and expect the upcoming crash to be economically and financially devastating to the tech and financial sectors. But I also have a pretty dim view of human intelligence, too, and see way too many parallels in LLMs as bullshit artists to humans as bullshit artists, too.

It’s that they are trying to use statistics to encode entire thought processes into hidden variables from conversation snippets. They want to use statistics to go from many individual interactions to a large model, and then use that model to predict individual interactions again.

Has it been shown that the human brain doesn’t model the world in a similar way, though? A huge portion of human knowledge is both stored and transmitted in the form of language. Lots of human knowledge also follows the garbage in, garbage out theory, where you can have entire areas of knowledge that aren’t actually true but might be internally consistent, at least within certain scopes: conspiracy theories, belief in the supernatural, entire academic disciplines built on a religion or theology that not everyone believes, etc. Or even world building in fiction, the words on a page can be enough to convey ideas such that it “tricks” human brains into filling in the gaps so that they internally see a rich, fleshed out world that is entirely fictional and where specific details might not find strong direct support in the underlying text.

it has no concept of correctness

But statistical weight on what is more or less likely to be correct still makes a difference to objective quality of the outputs. If the model weights are trained on the reality that high quality university texts describe something and reflect some sort of underlying model of what is described using language, then can’t the model itself learn as much as a human could from those words on a page?

All models are wrong, but some can be useful. And different models have different quality in different domains. So although I don’t believe LLMs will overtake the hump of getting ahead of human knowledge, I also don’t believe that any given LLM can be evaluated on quality, and that Facebook’s LLMs are significantly behind other LLMs we see.

And that maybe a huge part of it is its internal process of preparing the model to evaluate the quality of its inputs, such that the output it produces can also score high on quality.

At least they moved onto year-based versioning. That was probably the best part about the 26/Tahoe release.

Yeah, but that’s always been true of paid software licenses for a particular version: it reaches EOL and you have to decide whether to live with the possibility of unpatched known vulnerabilities or pay for an upgrade to a more recent release.

MS Office has been doing this from back in the Windows 3.0 days at least.

Edit 2: Nevermind. 13th October is the day Microslop has chosen to fuck me: https://support.microsoft.com/en-us/office/system-requirements/end-of-support-for-office-2021

You’ll still get a few years before the software becomes remotely disabled, though. This story about Office 2019 losing functionality follows Office 2019 losing support in 2023. If that’s the rate things go, then maybe Office 2021 will lose functionality either 2 years from now (7 years after release) or 3 years from now (3 years after losing support).

This is really, really cool. Makes you appreciate just how much this team does for reverse engineering what should be normally documented hardware.

This one is the same author addressing this specific Mozilla release.

briefly released millions of tracks that were scraped from Spotify via BitTorrent.

That’s just an awkward sentence construction but it makes sense: they released track via Bittorrent. The tracks were scraped from Spotify.

I sold my car that was purchased from a dealership via private party sale.

I charged my laptop that normally accepts 100W via a 20W phone charger.

I would’ve used a “which” phrase with commas to avoid the confusion, but the sentence as written is valid and makes sense.

Unless it can be paper thin this does not look better than magnetic tape.

As the article explains, the whole purpose here is to be able to store data on a medium that can endure harsh conditions, including heat, moisture, radiation, and physical abrasion. The company’s website claims the medium can retain data for 5000 years without power, and is water and fire resistant.

I reckon you could scratch it pretty easily.

The underlying ceramic film is already used for protecting tools like drill bits and saw blades from physical damage, which is why it was chosen for this project. They already found one of the most durable materials in the world, and asked whether they could store data using that already-durable material.

Reminds me of an AI tool that could identify authorship of articles with surprisingly high accuracy, and then they peeked under the hood and realized it was just looking for the author byline at the top of the article that says “By John Doe,” where it completely failed if the article didn’t explicitly say who the author was.

Motorola Mobility was spun off from Motorola in 2012 and sold to Google. Then Google sold it in 2014 to Lenovo, the Chinese company that had also previously bought IBM’s entire personal computer business.

Original Motorola, renamed Motorola Solutions, retained the rights to the Motorola name in everything except cell phones, and continued to manufacture radio and communications equipment and other signal processing equipment (including stuff like cable TV boxes). They remain a major contractor for militaries, law enforcement, and fire/EMS emergency responders.

If we’re talking about Motorola cell phones, we’re talking about the Chinese owned company, not the American owned company.

This operating system contains code known to the State of California to cause cancer or reproductive harm.

It’s funny. I see the phrase “AI doomsday scenario” and I immediately picture devastating cascading consequences caused by someone mistakenly putting too much trust in some kind of agentic AI that does things poorly and breaks a lot of big important things.

I’m just not seeing a scenario where AI causes devastating disruption based on its own ultra competence. I’m much more scared of AI incompetence.

Apple supports its devices for a lot longer than most OEMs after release (minimum 5 years since being available for sale from Apple, which might be 2 years of sales), but the impact of dropped support is much more pronounced, as you note. Apple usually announces obsolescence 2 years after support ends, too, and stop selling parts and repair manuals, except a few batteries supported to the 10 year mark. On the software/OS side, that usually means OS upgrades for 5-7 years, then 2 more years of security updates, for a total of 7-9 years of keeping a device reasonably up to date.

So if you’re holding onto a 5-year-old laptop, Apple support tends to be much better than a 5-year-old laptop from a Windows OEM (especially with Windows 11 upgrade requirements failing to support some devices that were on sale at the time of Windows 11’s release).

But if you’ve got a 10-year-old Apple laptop, it’s harder to use normally than a 10-year-old Windows laptop.

Also, don’t use the Apple store for software on your laptop. Use a reasonable package manager like homebrew that doesn’t have the problems you describe. Or go find a mirror that hosts old MacOS packages and install it yourself.

This write-up is really, really good. I think about these concepts whenever people discuss astrophotography or other computation-heavy photography as being fake software generated images, when the reality of translating the sensor data with a graphical representation for the human eye (and all the quirks of human vision, especially around brightness and color) needs conscious decisions on how those charges or voltages on a sensor should be translated into a pixel on digital file.

Cutting edge chip making is several different processes all stacked together. The nations that are roughly aligned with the western capitalist order have split up responsibilities across many, many different parts of this, among many different companies with global presence.

The fabrication itself needs to tie together several different processes controlled by different companies. TSMC in Taiwan is the current dominant fab company, but it’s not like there isn’t a wave of companies closely behind them (Intel in the US, Samsung in South Korea).

There’s the chip design itself. Nvidia, Intel, AMD, Apple, Qualcomm, Samsung, and a bunch of other ARM licensees are designing chips, sometimes with the help of ARM itself. Many of these leaders are still American companies developing the design in American offices. ARM is British. Samsung is South Korean.

Then there’s the actual equipment used in the fabs. The Dutch company ASML is the most famous, as they have a huge lead on the competition in manufacturing photolithography machines (although old Japanese competitors like Nikon and Canon want to get back in the game). But there are a lot of other companies specializing in specific equipment found in those labs. The Japanese company Tokyo Electron and the American companies Applied Materials and Lam Research, are in almost every fab in the West.

Once the silicon is fabricated, the actual packaging of that silicon into the little black packages to be soldered onto boards is a bunch of other steps with different companies specializing in different processes relevant to that.

Plus advanced logic chips aren’t the only type of chips out there. There are analog or signal processing chips, or power chips, or other useful sensor chips for embedded applications, where companies like Texas Instruments dominate on less cutting edge nodes, and memory/storage chips, where the market is dominated by 3 companies, South Korean Samsung and SK Hynix, and American company Micron.

TSMC is only one of several, standing on a tightly integrated ecosystem that it depends on. It also isn’t limited to only being located in Taiwan, as they own fabs that are starting production in the US, Japan, and Germany.

China is working at trying to replace literally every part of the chain in domestic manufacturing. Some parts are easier than others to replace, but trying to insource the whole thing is going to be expensive, inefficient, and risky. Time will tell whether those costs and risks are worth it, but there’s by no means a guarantee that they can succeed.

2 lanes in each direction with a middle lane? That’s a big chunk of Texas, especially when weighted for population.

The default in most other states is that opposite direction traffic on a divided highway don’t have to stop. The states differ in what constitutes a divided highway, but generally at least 5 feet of space or a physical barrier between the lanes would qualify. In Texas, however, there is no exception for divided highways, and the key definition is “controlled-access highway,” which requires on/off ramps and physical barriers between traffic directions, or “different roadways,”

So for a 5-lane road where there are 2 lanes going in each direction with a center lane for left turns, Texas requires opposite direction traffic to stop, while most other states do not.

Waymos were violating a Texas state law that requires cars to stop when a school bus stops, even in 2+ lane roads separated by a paved median, even for traffic going in the opposite direction:

https://liggettlawgroup.com/wp-content/uploads/2019/09/School-bus-laws-img-1024x657.png

The requirements for opposite side traffic in multi-lane roads is pretty rare and might be unique to Texas. And yes, human drivers fuck this up all the time, too, leading to a lot of PSAs in Texas, especially for new residents.

It’s bizarre how if you drove through twenty bus stops in three days, you would not only lose your license but be in jail on multiple charges.

This is a relatively unique Texas law that requires cars to stop when school buses are loading or unloading passengers, including on the opposite side of the road going the other direction. The self driving companies didn’t program for that special use case, so it actually is a relatively easy fix in software.

And the human drivers who move to Texas often get tripped up by this law, because many aren’t aware of the requirement.