Algorithmic patents amount to patenting maths which, by very longstanding precedence, is not a thing, for good reason. Same goes for business methods and other stuff.

In the EU there’s only one way to patent software and that’s if you’re using it to achieve direct physical ends. E.g. you can patent washing machine firmware in so far as you patent a particular way to combine sensor data to achieve a particular washing result. Rule of thumb: If, 30 years ago, you’d have an electromechanical mechanism to do the task then you can patent the software that’s now replacing it.

Oh: It’s also possible to patent silicon, that is, you can patent your hardware acceleration methods for video decoding. That doesn’t extend to decoders running on general-purpose hardware, though.

If you want to monopolise your brand-new hash algorithm there’s a simple way: Don’t publish the source, use copyright to collect royalties… though that doesn’t mean that reverse engineering is outlawed, especially if necessary for interoperability. Practically speaking nope hash algorithms just can’t be protected which is fair and square because it’s academia who comes up with that kind of stuff and we paid for it with taxpayer money. Want to make money off it? Get tenure.

DeepL has always used machine learning, and they already switched to LLMs for some language pairs – not rebranded ChatGPT, but their own stuff. They’re also quite open about the model not being perfect, they’re advertising with things like “blind tests show our results sound more natural than the competition”, “our model output needs fewer edits than the competition”, etc.

And yeah they definitely didn’t edit this one much from the English original. English sentence structure and American idiomatics all over the place, it’s tedious to read. Quite, but not entirely, as bad as this.

The vast majority of sales are made to US based firms so they likely have a lot of sway.

The sway is TSMC uses ASML EUV lithography machines and the US holds patents on those because they did foundational research regarding EUV lithography. Also, the EU hasn’t put China on the “it is illegal for EU companies to kowtow to US sanctions” list. Ironically ASML could sell to Cuba and Iran. If the EU were to tell ASML to sell to China the US would be free to not buy ASML machines any more and, doing that, kill off Intel’s fabs.

None of this stuff has military relevance, you don’t need or even want to use small nodes (which require EUV) in military applications you want hardened chips instead. Run off the mill consumer chips go all frizzy if an EMP looks at them sideways. This is about the US protecting US fabs, foremost Intel. Not the chip design part but the manufacturing one.

Europe hasn’t played the high-end end-consumer chip market for ages and I doubt we’ll do it any time soon. Having ASML, Zeiss etc. means that whoever actually produces that stuff wants to be friendly with us and strategically, both military and economy, our own production facilities are perfectly sufficient. Hence also why ESMC will only go as small as 12nm, it’s the most cost-effective node size and performance is perfectly adequate for a missile, a CNC mill, or a car infotainment system. Or the gyroscope chip in your phone (it’s almost certainly a Bosch), EUV doesn’t make a lick of sense when you’re doing MEMS. Where we have to catch up is chip design lets see how that RISC-V supercomputer chip turns out.

There are virtual spaces that are currently occupied by native inhabitants and deep pocket billionaires are literally bulldozing their way into the space

Nope. Usenet has been dead and conquered ages ago, others are up and on the rise (hey you’re posting on one) but mass adoption doesn’t seem to be anywhere on the horizon. Yes, the internet has changed after the Eternal September, the normies brought their economical system with them.

What I wonder, though, is what that has to do specifically with underpaid Kenyan data sorters or are you colonising their struggle with ours?

Fuck Spez

(Hey noone else said it in this thread so I think I have to)

Plenty of self-driving trains around, generally metros where frequency and 24/7 operation is a great boon to overall service quality – you don’t want people to look at schedules, you want them to go to the station knowing there’s going to be a train in a couple of minutes, tops.

It’s way different for long-haul service, freight, passenger, doesn’t matter. Longer and less frequent trains with way more passengers in them, and you probably need other staff too, like someone needs to run the bistro. The tracks they’re running on are also way less predictable, with a metro you can have station screen doors everywhere (which btw necessitate automatic driving, humans aren’t accurate enough) try that with an international train: Regions much less countries can’t even agree on uniform platform heights. Much less door locations: Automated long-haul would require dedicated platforms at every station and while those could be served by trains with drivers, trains nowadays are all smart enough that including a button “stop at exactly that location, to the half-centimetre” isn’t an issue, those trains would have to have doors at the right location. Now go ahead and convince Germany and France that they need to replace all TGVs and ICEs to have doors in the same location as your regional trains.

Oh and none of that automation tech used with trains uses machine learning, btw. At least not at the basic level, when it comes to actually driving the train. I do remember watching a documentary about Singapore’s metro, where they have an ML algorithm scheduling track maintenance, minimising not service interruptions as such but impact on people’s commute. First the workers complained that none of the orders made any sense, then the developers made the computer spit out context and motivation alongside with the orders, workers changed their tune to “that’s fucking brilliant”.

…which, actually, brings me to the conclusion: Also with automated systems we’re going to need maintenance which isn’t going to be automated any time soon. If you automate a metro that currently doesn’t run 24/7 you don’t have that many drivers in the first place, and probably have other jobs for them to do. Automating really is about making “a train max. every five minutes, 24/7” possible without breaking the bank.

Very relatedly: If you’re a EU citizen make sure to sign this initiative.

If not a EU citizen do not sign it, that’d be the opposite of helping, but do enjoy numbers go up from afar. Also you can spread the word that’s fair game.

The only reason I can think of to get that data is to circumvent archive.org’s limitations on renting out stuff so your bots can slurp books without getting throttled.

That is, my best guess is that this was pirates seeing if they can’t add to libgen.

Or it’s a “prove you can do stuff” kind of deal.

ROMs back then got erased by UV light, EPROM. EEPROMs are a bit newer (though still ancient) and can be erased electronically, nowadays it’s a very sane idea to just throw flash storage at the problem. I think you can get modern replacement for pretty much any ancient form factor.

The way those things are used are basically big logic tables: Instead of using a bunch of logic gates, you store the output that’s expected given a certain input. Completely ancient technique, the limiting factor is storage space and sensibility – storing all addition results of two 32 bit numbers uses a lot more transistors than a 32-bit adder, but if what you want to put in there isn’t a thing that can be implemented few standard TTL components throwing storage at the problem makes sense even if you never plan to reprogram it because burning a custom set of transistors onto silicon is expensive.

Some more technical info.

It’s an legit 8-bit CPU implemented with TTL chips, what makes it a different beast than what they did back in the days is that its microcoding isn’t kneecapped. It would absolutely have been possible back in the days to build exactly such a thing, even from precisely those components. At least the TTL part, that is, I bet there’s wibbles around VGA etc. And because I already hear the detractors yes, 8-bit CPUs were microcoded: They decoded single external instructions into a stream of “load from memory, fetch from register so and so, switch on the ALU, put what’s in the ALU output somewhere”. They kept it as simple as possible and it wasn’t reprogrammable but that stuff there, that’s microcode.

Implementing CPUs in TTL chips also isn’t a new idea, that’s how early minicomputers were made (later on they got some specialised chips). And those things also used ROMs for their microcode. So you could say that this is a minicomputer capable of pretending to be different 8-bit microcomputers.

FPGAs are a completely different technology, those allow you to arrange logic in a (more or less) arbitrary topology. That is, looking at that board with all those TTL chips, it’d be the equivalent of being able to re-route all the board traces as you please.

He’s accused of tolerating such distribution even after being alerted of the fact so, yes, he’s accused of distributing it. For an online platform “we didn’t know about it because we don’t moderate proactively” is a valid defence, “authorities told us about it but we didn’t feel like doing anything about it because can’t be bothered / frozen peaches” isn’t.

You can’t automate generation of shape keys. An artist needs to go over every single asset and make it work for every single extreme point on every slider, then make sure that the automatically derived in between points look good and fix those if required, in all slider combinations.

And it’s probably still going to clip during some animations because going over absolutely everything is just prohibitively expensive.

Why is the term “Body Type A” and “Body Type B” present at all when there are clear pictures of the two options that speak for themselves? It feels like just going out of the way to include “the corporate approved buzzwords intended for maximum synergy with the brand!”

“Type A” and “Type B”, I assure you, are not things corporate or marketing came up with. This is programmer speak for “I don’t want to name it but can’t call it foo and bar either because normies will be seeing it”.

As said: This is a re-release. The game and its assets was originally never designed to support anything but a strict binary, but the pronoun vs. body type thing was trivial to do, so they did it. And then for some reason avoided “male” and “female” because face it that sounds like a good idea especially if you’re not overthinking it and the labels were left in because probably also easier to do. Or just didn’t consider the alternative.

That is: You’re assuming intent when there’s simply economy of action. You might call it laziness, but then the people who did that release had 10000 other things to do besides that.

Putting in half a dozen body types and a boob slider shouldn’t be a ton of work

Body types no but you also need armour and clothing for everything. You quickly get a combinatorial explosion which you can then reign in with shape keys (“sliders”) which make all assets harder to develop.

That’s morph targets and you just increased the budget for the character model and every single set of clothing and armour by a whole magnitude. Might even influence animations, though I guess with Elden Ring being the game that it is those are the same for everyone.

Games that do this aren’t being progressive or inclusive, they’re changing the color of the cup that my drink comes in and pretending it’s an entirely new beverage.

The thing is… if you use “dude” and “chick” in the body type descriptions you’re implying gender identity. There may be better options that “Type A” and “Type B” but dude and chick ain’t it because it simply means male and female.

In a very flexible system, you could use more granular options like “wide shoulders”, “wide hips”, “boobage”, etc, to freely mix+match everything. It’s also expensive to develop and even more expensive to create clothing for and a gazillion times more expensive to make really good-looking clothing for (fabric folds and flow aren’t easy). From a developer’s perspective, looking at the work involved really makes you want to say “We’ll just tell the player they’re now Geralt of Rivia and that’s it”.

I think for most games the appropriate choice would be to have an early radio button, saying “male/female/it’s complicated”, the first two options hiding every enby option including pronoun selection. That’s right-out trivial to do and just good UX. And yes the body types should be called male and female, you already selected “it’s complicated” so it’s clear that when you’re selecting a body, you’re selecting a body, not identity.

As to laziness: Eh. Noone’s going to start a research programme on how to do things in an optimal way for a re-release. Someone had a look at the code and assets and thought “hey we can support separate pronouns and bodies without doing anything more than providing an option” and that’s exactly what they did, using the extent of knowledge and consideration that was already in-house. Yep, it very well can happen that if you take your foot out of one thing, you put it right into another.

As to “primary/secondary”: One of the options has to be to the left, or on top, of the other. Ain’t no way around that. I mean you could put option B on the left of option A to cancel things out but now you’re being confusing. More importantly you can make it so that none is selected by default.

Am I onto something or is this all crazy talk?

Yes and no you’re being quite personal, and I include your perspective shift into the POV of others in that, about things that will never make 100% of the people 100% happy because technical reasons. The perfect is the enemy of the good and all.

I’m not sure if you could call Talos Principle indie. Croteam is an ancient company (of Serious Sam fame), they sold out to their publisher some years ago (Developer Digital). Wikipedia says 42 people, that’s about the same ballpark as Wube (Factorio), way smaller than Coffee Stain (123), which yes vibes heavily indie (Goat Simulator!) but is part of Embracer Group.

If you look at Developer Digital and Embracer group they’re not really that small – certainly not smaller than CDProject Red, which is very much throwing AAA money at their projects and definitely had their own big business culture fuckups. They’re simply more distributed, instead of orchestrating one or two big projects they have multiple studios working largely independently on small to mid-sized projects. Talos and Satisfactory are AA scale.

Is Wube indie? Well, at least at the start they were, growing with Factorio’s early access. Still independent, as far as I know. Budget-wise they’re certainly not operating on a shoestring, though… you also have to take into account that they’re taking their sweet time for everything. Also AA.

A would be stuff like Celeste. That’s a broad category, I wouldn’t really call anything B unless you don’t have separate coder, writer, sfx/msx and gfx. Maybe toss the writer but anything under that and you’re smaller than minimum demo group size.

All this is to say: Can we please stop dividing the industry into “AAA” and “Indie”. CD Project Red is independent. They’re doing AAA. One is budget, the other is whether the studio has a corporate overlord lording over multiple studios. Game quality is a third measure. System requirements yet another: Factorio has no issues melting your CPU even though it’s highly optimised, then you have B-budget projects which melt your box because the dev has never heard of polycount and a background prop toothbrush has 400 quads… per bristle.

AI image generators don’t “consult” source images to generate an output.

Well, you have an artist breaking things down for an audience understanding neither the technical nor artistic aspect…

Modern AI generators are increasingly good at generating text. They still struggle a bit

I mean… SDXL still struggles a lot. The only thing you can get it to spell reliably is probably “Hooters”. There’s the one or other lora which makes it not suck completely but it’s still nowhere near actually good at generating text, the training just isn’t there. And even with that in place things like signatures are probably going to be gibberish.

While a naive (and cheaper) approach to AI generation doesn’t use layers, there are generators which do use layers,

Unless you start off training by feeding the model 3d data (say, voxels) alongside 2d projections I don’t think it’s ever going to develop a proper understanding of these kinds of things. Or, differently put: Learning object permanence (of sorts, related) is a meta-cognitive abstraction step that just won’t happen with the type of topologies we know how to engineer. It’s probably like 90% on the way towards AGI, so to get a simple topology to understand it we have to spoon-feed it permanence information alongside the (apparent) non-permanence.

Satisfactory did it earlier.

It’s quite easy to trick people with untrained eyes… for one, they have no idea what “consistent illumination” and stuff means. And something being off doesn’t mean that an AI made that mistake because humans make mistakes, too – photographs don’t, but the general problem is not just about telling realistic stuff apart but also illustrations. You’re looking specifically for mistakes that AI is likely to make, but humans are practically never going to make. And yes humans get hands wrong all the time.

Here’s a good video about what to look for and what not.