Well as you said FRAND is one reason private companies would fund research and then donate it to a standards body, and a “good” one.

Not the only reason though. A lot of things are driven by adopting standards. If you make your very own technology and other companies need to adopt it they tend to want to deal with standards not a single source product, so you get wider adoption by publishing “your thing” as a standard. It does basically burn most or all of the money you invested R&D, but in exchange you get to enlarge the market. You also get to be the first company that can supply a product that uses the standard. So investing money into a new radio strategy and having that be part of the next WiFi standard you are trading that R&D money for being one of the first companies with the “new” WiFi in your basement station (or sort phone, or whatever).

You also get to shape the future direction of a market. So for example (sticking to radio technologies again) if you contribute a low bandwidth low power long range mode for WiFi you can make WiFi more applicable for things like location tags.

SANDisk I think makes FLASH but not RAM, so from their point of view high bandwidth FLASH displacing some HBRAM demand is a huge win for them, any hardware that decides to use this and less HBRAM makes the market SANDisk participates in larger. SKHynix doesn’t get that same boost because I’m pretty sure they also make RAM. SKHynix however gets a boost because unlike say the other 15 companies that will eventually make high bandwidth FLASH SKHynix (and SANDisk) will be making it from day zero.

Yes, there are long running complex lawsuits about all manner of things you might think are intuitively obvious, however just because the last lawsuit was long running and complex doesn't mean the next one about the same thing will be. In theory if the prior court found that header files are public interfaces and don’t have copyright protection the next lawsuit involving copyright on header files will have the defense lawyer point at the prior judgment and saying “we don’t need to litigate this, it was clear two years ago, and nothing has changed to make this any different, we should do that again” and the judge doens’t have to do the exact same thing, but most will, unless the other lawyer has good counter arguments why it is different in this case (and NOT arguments about the original judgment being incorrect). Sometimes the “good arguments” are merely “that case was decided in Californian and we are in Texas, and the laws are different here!”.

Legal matters can be counter intuitive, but more frequently they actually make sense. Early cases about things are complex and long because the legal process tries very hard to make sure they have a good answer, and that they clearly explain the reasoning behind the answers they give, and the limits of the answer. The results of legal battles that reach a verdict are not just a judgment about how a clash between two parties will be resolved, but about how all similar cases should be resolved. Think of a verdict as something that has cached results (but in a system that can do fuzzy matching on cache lines). A settlement on the other hand produces a result between two parties but no new cache entries.

I’m a lot less sure. I would think a lawsuit could start based on object code, and during discovery get source code and continue if the source code looks similar enough. Doing a whole lawsuit on object code would face significant issues convincing a judge/jury that object code is “expressive enough” to manifest copyright. With the defense arguing that wholly independent implementations of something compile down into the same (or substantially similar) object code.

It should be “easy enough” if you think you have some highly similar object code isolated to convince a judge to allow a civil suit to start and to compel discovery of the source code so you can examine how similar actual code (which has already faced court scrutiny about weather or not it is expressive enough to receive copyright protection) is. Even there you can get some suprising results. Like can a Google Employee who was previously an Apple employee and implements the same function at both companies in basically the same way commit copyright infringement? (In something like a topological sort where the algorithm is a known art, it is entirely possable that writing it from scratch for the Nth time produces a line for line identical copy including variable name choices and the bulk of the comments even if the author doesn’t actually consciously remember prior version of the code!)

Just because code quality improves doesn’t mean the code looks like what a human writes. Currently AI code seems to not favor writing lots of functions, do it’ll basically expand out the same code in multiple places. Which I admit is also what jr programmers tend to do and maybe you can quality that as something that is code quality and will improve. You can frequently use a prompt to make it “don’t write the same code multiple times, define functions for anything repeated (maybe over a specific complexity)” & that reduces that issue.

I’ve noticed AI generated Swift tends to not use default arguments much, almost never uses custom operators, and uses unnamed method arguments way more frequently then most human written code. Sort of. A lot of example code and people showing things in forums and doing toy programs “in public” are similar which makes me guess those end up in training data sets more then signifiant works that do most of that stuff. Or I write atypical Swift, and have worked at places that do the same (in part because I get to set the coding standards at some of those companies!). At any rate I think _some_ AI code has “tells” which is largely independent form bug rate if not actually “code quality”.

As a practical issue commercial software is almost always closed code, so nobody can see the source and simply search for anything that might share code with anything they have copyright on. The open source code isopen. So if you have anything copyrighted you can search the open source to see if anything looks like it is worth filing a lawsuit over. Especially if you have a competing product. So if you are say Intel you can check to see if any of gcc is similar to Intel’s commercial compiler and sue claiming any recent declines in sales are directly related to gcc’s “blatant” infringement. Or if you have a non-compiler you can sue and just complain about common stanzas of code but not establish specific damages so it might be a less valuable lawsuit.

You could even sue if your closed source code got that line from AI as long as the employee that the source code control system claims they wrote that one themselves, and as long as your source code control system says that line was written before the open source project had it’s equivalent code (or if your lawyer doesn’t think to check those things, and the OSS’s lawyer doesn’t think to raise it in defense), so at least chances are rare that the OSS code invents the line and commercial closed source steels the line (via AI) and then claims the reverse happened.

Winner! Winner! Chicken Dinner!

Yep, even when AI generated code is human reviewed to determine that the code does what it looks like it should, and has no subtle issues (or only very very subtle ones!), you still have the huge legal issues (the exact ones you identified).

Depends on how solvent NVIDIA thinks those companies will be. Also I think creditors frequently get paid earlier in a bankruptcy process (before investors at least). Maybe they have it structured like a mechanics lien where they get 100% of the loan back on bankruptcy (I don’t know if that is actually possible), or if they just huger that in a bankruptcy they will get about 80% of the loan back, so they want to make the loan as far ahead of bankruptcy as they can so they get a bunch of payments followed by bankruptcy and then they get a chunk of money awarded by the bankruptcy court (and/or maybe return of the assets?)

Ah! I think that is the one! I bet they are structured as a secured loan, so on bankruptcy the bankrupt company can either surrender the assets (the hardware Nvidia loaded you money for), and Nvidia keeps all prior payments. OR they sign a new agreement both the company and lender agree to which is then omitted from the bankruptcy discharge (i.e. the new agreement can’t be washed away by bankruptcy, so if you still owe say $3m Nvidia can stand firm on that price, or even knock a little off, but whatever is agreed to the bankruptcy court leaves it to stand -- normally, sometimes they will decide there is some sort of bankruptcy fraud going on and you are sheltering money with something like that, normally only if family or close friends are involved though), or the company going into bankruptcy can pay Fair Market Value for the property.

Nvidia might get screwed if companies going bankrupt decide it isn’t worth salvaging the AI hardware either because they pivot to a non-AI busness, or at least a busness where they don’t need Nvidia hardware (like they rent “AI compute” from Google or something, or only needed the Nvidia hardware to refine models and they decide to just run the models they already have and not refine anything).

Nvidia also might get screwed if lots of companies decide AI is BS and the fair market value for AI hardware declines.

Nvidia is also screwed if the whole AI market implodes for a similar reason.

Still it is pretty easy to see how Nvidia is taking a risk, but could “earn” way more money this way. Plus Nvidia gets the interest payments, not a bank (then again Nvidia is taking the risk, not a bank). I don’t know if it is a good idea or not, because it hinges on something I’m awful at predicting: timing. If the house of cards comes down soon Nvidia takes a big bath. If the house of cards stays up for a while Nvidia breaks even, if it stays up long enough Nvidia makes the even bigger bucks.

Actually it is debatable. Anthropic is not making money, they are burning through VC cash. OpenAI is not making money, they are burning through OPM (“Other People’s Money”) as well. NVIDIA _is_ making money selling the AI hardware to the various companies pouring money into finding an AI model+business model that lets them make money. SpaceX’s xAI isn’t usefully broken out into it’s own P&L, but if you want to bet there IK’ll bet they are losing money at the moment as well, if you take the bet and lose you owe me one fancy Starbucks coffee and pastry, if I lose I’ll buy you 5 shares of SpaceX (that is around $1000 vs your $10). Did I forget anyone? Palentier? Also OPM (VC at the moment).

I doubt Apple or Google are making money on AI either, although it is more debatable because they have a bunch of products they can bolster sales with it (people buying a new iPhone because they want something they saw “Apple intelligence” will bring, totally forgetting that Apple hasn’t delivered on their AI promises from 2 years ago yet!), or Google may sell more smart speakers “powered by Gemini” then they sold of their prior assistant powered speakers, or get more Google searches because of Gemini answers. Although I expect the searches will merely manage to better hold onto market share by offering Gemini, which is important, but defending a multibillion dollar a year Business isn’t at all the same a “making money with AI”. Google has definitly said they are cutting headcount by using AI, but I’m not sure if they actually have a productivity increase or merely an increase in code per unit time even if the volume says nothing (or is of lower quality so it produced more bugs per unit code, and more critical outages per unit codebut more volume of code fixes!)

I’m not saying that NVidia are the only ones that can ever make money at this, or that only the hardware makers will ever make money, but that is how it is right now for sure.

As I understand it Mythos’ “big leap” is not in finding specific flaws it is in chaining them together into a “bigger” flaw. So finding a minor issue in curl that lets you put a file where you shouldn’t, plus a flaw in something that assumes some file location is “safe” and it doesn’t have to parse things with an advassery in mind, plus a flaw in something that relies on that thing, and so on.

When doing that kind of security work you don’t need to find a bunch of significant flaws in each tool, just a minor flaw in places that turn out to be useful when combined with say up to 9 other minor flaws. So from the viewpoint of cUrl which doesn’t rely on a lot of other tools to provide its services nothing has changed. The pain is experienced on a wider scale like over a whole OS where there are a lot of tools any of which might contribute a minor flaw so Mythos can find way to gain “the prize” (maybe remote execution, or a privilege escalation, or both).

Maybe a better way to think about Mythos is it doesn’t have to hyper focus on one tool like “can I break into the system using cUrl?” (and is not actually any better at that question then prior AI), but it does a far far better job at answering the question “can I break into the system using up to a dozen or so flaws together out of this pool of 1000+ tools?”. I assume it may be a bit better at finding flaws in single tool if the flaws require putting more bugs together or more steps to reach the state where an existing flaw shows up, but again that isn’t the big deal. The big deal is at a system level it puts multiple sub-critical flaws together to combine into a critical flaw. (queue transformers joke here)

Is Cuda a lock in because there is a critical mass of solutions written in Cuda and people that think about problems in terms of Cuda already so nothing is really going to unseat it that isn’t a close clone of Cuda and making one of this is for some reason impossible, or is the problem that you can make something else that lets you be expressive in the imprint ways Cuda is while giving the backend the same kind of flexibility to schedule operations, but nobody else has made one that isn’t “too buggy” to use on real world problems?

In the distant past very few C/C++ complies existed, they were “too complex” for small companies to make, and now we have very few commercial compilers and a billion open source projects that are all forks of gcc or llvm (or a fork of a fork of the llvm derived clang). We don’t have a billion non-C derived programming environments though (we have a few, JavaScript is popular, and I’ll argue Java is C-derived, although removing pointers form C doesn’t leave a lot, so I’ll also accept it as a distinct environment, but if so, so is Swift, and Rust also counts as distinct...still that is only handful). It doesn’t prove a lot, but I would say even the moat of a programming language and environment only lasts so long.

Do you actually believe that? I mean, yeah sure “we asked them what was up and they gave a flimsy excuse” doesn’t mean you have to believe it!

The only thing that points towards them maybe telling the truth is it might be obvious if the data center were operating and you don’t want to get caught in a provable lie. However it is also possible the data center is partly operating while construction continues and they figure “hey there aren’t people coming and going, who will know if the data center is operating as opposed to testing equipment if we get caught!”.

e benchmarks I saw had it about the same. The 2020 M1 Air slightly faster in single threaded, slightly slower in multithreaded, or do I have those two swapped. Either way, it's abouth the same overall.

Geekbench puts them at Neo 3535/8920 (s/m) & MBA 2347/8342 (s/m). So the Neo is significantly ahead on single core performance, and ahead (but just barely) in muti threaded even with the reduced core count! Which is decent for half the price! (well if you get the EDU discount on the Neo, the M1 MBA doesn’t have a EDU discount that I know of).

To be clear at the price point not too much slower would be Ok, but the benchmarks have it at way faster for single thread, and faster but basely a tie for MT.

To be honest a used M1 Air or M3 Air would seem a better deal

A used M1 Air at under the Neo's price would be a good deal. The M3 at the Neo’s price would be a great deal. I mean the Neo is pretty damn good at its price point. It is fast, it works surprisingly well for its RAM configuration. The +$100 model has touchID and slightly less pathetic local storage. The Neo’s display is physically smaller, but the whole device is quite small, some people prefer smaller devices for carrying about and use in cramped areas. I mean from my view point (16” MBP with large external displays) both the Neo and MBA have tiny cramped displays! However both are shockingly fast for their price.

The MacBook Neo gets a fairly high repairability score. Most people who have disassembled it seem to be of the opinion that since it isn’t going for absolute minimal size and weight they used very few adhesives and lots of screws. So it is pretty simple to take apart and put back together. Apple does also make “self repair” kits for many products amiable to rent with an unreasonable deposit (purchase also available, but not useful to most people), but has apparently decent instructions and such to get things done.

As for upgradability, nope, they are headed away from that as fast as they can. No RAM upgrades on any modern Apple device, the RAM chips are wire bonded to the CPU, which at least means they use lower voltage swings and get somewhat better latency out of the same parts. Not in general a tradeoff I would make (I would rather have DIMMs and be able to do a late-life RAM upgrade to get more useful years out of a device rather then be stuck at my purchase RAM allotment forever -- and/or buy a low RAN model from Apple and do a day 1 3rd party RAM upgrade). To be fair to Apple customer installed RAM, and factory installed RAM that managed to work loose were the number one and number two repair issues for their upgradable devices (or maybe just laptops?) prior to starting to solder down RAM. Which statically means a shit ton of people thought Apple just made crap computers that flaked out at random and never brought them in for someone to tap on each DIMM and “fix it”. So soldering the RAM down decreased warranty repair costs, decreased out of warranty “customer comp” repair costs, and increased perceived reliability amongst people that don’t take flaky laptops into an Apple Store and try to get someone to look at it.

The obvious downside is I pay more when I buy a Mac either because I don’t buy enough RAM for the full useful lifecycle of the CPU, or because I do and Apple charges a lot of money for it (well until recently, due to long term supply contracts Apple’s RAM cost is very low, so the normal 50+% profit margin they take on RAM now seems highly competitatave with spot RAM prices)

The Neo is nerfed in way to reduce cannibalizing the Air. It 2020 M1 CPU performance with a smaller screen and slower I/O.

Or alternately it is intended to be a replacement for keeping the 2020 era M1 MacBook Air around for “special retail partners” (i.e. Walmart and Costco) to sell at $799. The MacBook Neo other then the I/O much faster then the M1 MacBook Air, and in various “whole system” benchmarks apparently the slower SSD doesn’t really impact it too much. Which blows my theory that the M1 at 8G could get so much done without feeling slow because the SSD was fast enough that it blunted the effects of doing VM page-outs/page-ins.

The Neo will make money, but that will be largely due to growth in the educational ecosystem. Maybe some other Chromebook niches.

It seems very popular for people who “always wanted a MacBook” and never could convince themselves they could afford one. I don’t know how many will be repeat buyers (let alone climb up the product chain to the Air). If they are satisfied with the Neo they don’t need the next Mac to be faster. If they are not satisfied with it, will they blamee that on “Macs suck!”, or on “The Neo is just entry level, I need to spend 2x to 3x to get a faster one!”.

I do see a ton of people talking about getting one as a first Mac, or having bought one and asking all the “how do I make it do (some random thing they miss from Windows)” questions.

Enjoy your retirement TIm Apple, you nauseating man

Except he isn’t retiring, he is stepping down as CEO and assuming the Chairman of the Board position where he will be specifically responsible for interfacing with world leaders (i.e. more sucking up to the orange man). Which is nice for John Ternus I guess because he can run Apple without dealing with(directly) with Trump, and let Tim deal with tossing golden trinkets at him and assuring him Apple is bringing manufacturing to the USA while talking to every other world leader and assuring them he is bringing manufacturing to them, and to China and tell them manufacturing is going no where and all the other things are just tiny slivers of inconsequential volume to appease random world leaders.

So no, his replacement isn’t kissing any rings, Tim gets to keep kissing the same rings. I don’t know if he enjoys it (I suspect not), or just thinks it is his responsibility, or can’t liquidate his stock fast enough to deal with the fallout if Apple stops being favored by Trump (Tariff bypass) and starts being targeted (special Tariffs, and more levers that a President could pull...like having someone hand inspect each iPhone before it comes through customs which would move a scheduled launch day back by weeks, months or years depending on how slowly the inspections are carried out).