Expat here -- yes, that's a real problem. There are even well-located commercial buildings in central Lisbon that are decaying while things like that are sorted out. I honestly don't know why they don't do short-term (as in, three year or so lease) rentals while things are sorted out. Perhaps protections (regulatory or customary) in commercial leases that make it hard to get people out once an agreement has been reached between the heirs?
> This high failure rate is the reason for the sale, with upkeep made challenging by the fact that Cheyenne "is currently experiencing maintenance limitations due to faulty quick disconnects causing water spray."
I've been running my ML stack on Arch for the last six months and haven't had any issues so far. So perhaps not a big issue on a more stable distro like Ubuntu.
Absolutely. The framing around this is mostly about "when is the next one", but given the unlikely combination of errors and circumstances that led to this one being discovered, why would we think that it's the first time? It seems likely that this has been going on for some time, against various projects, and this is just the first to be noticed.
I got the impression at the time that stuxnet was Israel (albeit likely with US involvement). Has it been shown to be the US for sure? Or are there still competing theories?
Not from Lago, but from my own experience I'd guess it's due to the requirements to set up payroll in different countries. If you hire someone, even as a remote worker, in a particular country, you need a payroll processor to make sure the employee is paid with the correct tax deductions, pay those taxes to the local government, and so on. You also need to make sure that you comply with local labour laws. Doing all of that is far from trivial, so there's a high cost to taking on your first employee in any given country.
Intuitively you might think that the EU, with all of the freedom of movement laws, might make this unnecessary, but it doesn't. At least as far as I know, it's no easier for (say) a German company to hire someone in Italy that is is for them to hire someone in Japan.
Also worth noting that while there are "Employer of Record" and "Professional Employer Organisation" businesses who essentially have branches with payroll etc all set up in a bunch of countries, and can hire people there and act as the nominal employer while the employee actually works for another company, but at least the last time I looked they were very expensive -- at around thousands of $/€/£ per employee per month.
I was working for a company in France, while based in Latvia, and we worked according to B2B contract and I was just invoicing them and paid all my taxes. No problem.
True, that can be a useful solution. But you have to be careful; when I was running a UK-based company we had to get lawyers to carefully draft the contract so that it fulfilled the requirements to be truly B2B and not be something that would be deemed employment. I would expect the rules around that kind of thing to be even tougher in France, if anything.
An interesting article, with many good points (I particularly like the reminder that unless you're competing with OpenAI, AI is in the solution space) but despite the repetition that this time is different to the Internet, I do feel that if you replaced "AI" with "Internet" you'd get something that could have been written in 1996.
The core difference is that the internet was about distribution while AI is about intelligence, which includes the ability to create things.
The internet is probably the closest the comparison, which is why I get your point, but far off in terms of the scale of the implications.
The hardest thing about AI is that we don't have intuitions for it. It will make the exponential growth that the internet brought look linear by comparison.
I think the closest comparison to AI is the assembly line. None of the current AI offerings make sense without scale. AI also helps enhance some work, but it's not the product itself that the end user wants.
I'd say AI is more like a 3D printer than an assembly line, and not just because both sound futuristic.
Assembly lines make one thing efficiently, by breaking down the complete task into small units such that no single person needs much training. This is basically what UNIX has been doing since at least the command line pipe operator was invented.
3D printers are general purpose and can kinda make anything in principle, but in practice the cheap ones are very limited while the one for building a rocket engine is priced at "hire a team of full-time engineers to build and maintain it", and in any case these are not a good choice for reliable mass production. This seems like a better, though still flawed, pattern match to AI.
What quantization were you using? I've been getting some weird results with 34b quantized to 4 bits -- glitching, dropped tokens, generating Java rather than Python as requested. But 7b, even at 4 bits, works OK. Posted about it earlier on this evening: https://www.gilesthomas.com/2024/02/llm-quantisation-weirdne...
That would help, but I can imagine that on larger planes, especially if people connect their phone, then later their laptop, and so on, you could hit the limit of a 192.168.x subnet even during a flight.
reply