If they were archiving digital versions so they could be read by thousands in the future, this would be a totally different conversation.
Unfortunately that would be illegal.
boonhet
joined 1 year ago
It's a mixed bag.
The major companies that use it, contribute back to it (think Netflix, Sony, Apple, etc). If they didn't have a copyleft option, Sony and Apple would've built completely closed-source and wouldn't be contributing to anything. Netflix is different in that GPL wouldn't force them to disclose their modifications, they just preferred FreeBSD because of its networking stack, they would've probably just gone with Linux if FreeBSD didn't exist.
Of the ones listed, Apple and Sony have definitely contributed back to FreeBSD. Apple hired the lead LLVM dev and paid him to keep working on it. Sony has paid for LLVM development in particular, but also development of other parts of FreeBSD. Source on the Sony bit, which also mentions that for example Netflix is an even bigger contributor
There's plenty of benefit in contributing back to permissive projects you use, because you have less maintenance overhead if the diff between your fork and the upstream source is smaller.
Well, Sony contributes, but not as much as Netflix (mostly by paying for development, especially LLVM development)
Apple hired the founder of LLVM and paid him to work on LLVM.
LLVM is pretty important to FreeBSD.
Plenty of big companies contribute to BSD financially or let their devs commit code to BSD because maintaining your own fork is tedious work and the smaller your changes from the upstream source, the easier your life.
In my country for liability insurance at least, national traffic insurance fund website has a comparison tool. Just enter the data and get results for all the companies (minus the banks for some reason, they also offer insurance). The difference can sometimes be 5x or more and they offer the exact same legally mandated coverage (some have minor bonuses like "moose insurance" or legal aid, and no, those aren't even always the most expensive ones).
You start it remotely if you don't have a preheater. Not that idling is great for a car engine, but at least you'll have clear windows by the time you get in the car.
In the summer you can use it to have a cold car to sit into.
Remote climate control doesn't work on ICE vehicles without Remote start or auxiliary preheater.
My phone has access to my bank app, my car does not.
Can't find any near me, I'd have to fly to another country to get one that's not full of rust holes and decommissioned due to failing safety inspection which doesn't allow for perforation.
It's not necessarily that any of the open models are cheaper to train (for similar model size), it's more that China has deeper pockets. And MoE inference is cheaper than dense models I believe
The people who buy new vehicles usually do it for the warranty that lasts several years so the cost structure of ownership is entirely predictable.
You can also buy a 25 year old Volkswagen 1.9 tdi. Some years your cost of ownership will be fuel and oil changes (and those cars basically take no fuel), but at some point you'll need new unit injectors costing 2-3x the car's value for a set. Will it be the month after you buy the car, or will it be in 5 years? It's a lottery and you're better off replacing all 4 when it comes to that, rather than doing one at a time.
Cars in the range of 5-15 years of age are even worse because they're more complex and some of the parts are super expensive. And it takes around ~10-15 years to know if an engine or transmission design is truly reliable (if it has big problems, we might find out sooner).
I'm going to use German cars as an example again because those are the ones I know best, but for example when BMW had the M57 diesel, that was basically the best modern diesel engine. 500k km? 700k km? No problem whatsoever. Only real issues would be the occasional injectors, turbo and maybe some coolant hoses, but those usually started surfacing after 300k or 400k km, at which point you expect SOME issues.
Then N57 came out as its replacement and a few years later we knew about the potential chain and valvetrain issues. Kinda sucks, because they'd also moved the chain to the rear of the engine, but you can get it done for a few hundred euros yourself, or somewhere between 1500-3000€ at a shop. But it took nearly a decade before we knew about the crankshaft issues and those are an engine killer. I'm talking a full rebuild of the bottom end with a NEW, not re-ground crankshaft since these have a special coating from the factory IIRC. Prepare around 10000€, which incidentally could be more than the car is worth by the time you find yourself doing this work.
Now that engine's replacement, the B57, has been out for 8 or 9 years and as far as is currently known, it has nearly no typical issues beyond some plastic coolant pipes that get brittle over time. But we don't know what other common issues might start surfacing as more and more of those engines reach high mileages.
Now if what you're thinking is "just buy a Toyota and you'll have no trouble", well, Toyota has made absolutely horrible engines in the past, for example the 2.2 diesel. Head gasket failure was notoriously common. Again, something that you wouldn't necessarily know when you're buying a 5 year old car because the issues haven't started surfacing en masse yet.
The AI which cures cancers is architecturally pretty similar to LLMs in that it's based on transformers and GPT in particular is also based on transformers (that's the T in GPT). It is also generative AI, you give it tokens that describe proteins and it generates folding structures. It's like an LLM except the tokens aren't words or syllables like in LLMs.
It's used very differently and the energy cost is a few orders of magnitude less (you can run it on a single H200 with 141 GB VRAM for ~9000 tokens or an A100 or H100 for ~5000 according to this university and it even has a slow CPU/IO heavy first stage preparing everything so they can minimize the GPU time in stage 2).
There is yet hope that such AI applications will eventually benefit from all the research that's been put into LLMs. Obviously not everything will carry over, but there may be architectural improvements to be made in research models too.