Right. We've had mass data collection tools for a long time, web trackers, CCTV, microphones etc.
But what AI brings to the table is how much easier it makes it to combine and analyse all that data, build peoples' profiles. It scales infinitely better than hiring human analysts to track someone. It's too cheap and too tempting.
It'll still trickle down to people using the system and waste people's time (from development to the people working to produce all the things used by these companies and mopping up the impact this has to eventually the users) but that would definitely resolve one of my concerns
You can also very trivially do (codepoint | 1).leading_zeros(), then you can also shave one byte off the LEN table. (This doesn't affect the result because LEN[32] == LEN[33] == 1).
Piledriver (2012) has introduced TZCNT. LZCNT had already been supported by all AMD CPUs since Barcelona (2007).
Moreover, LZCNT is the more important instruction, because it can replace TZCNT with the addition of a couple of instructions, without using any branches, even in the more rare cases when TZCNT is desired instead of LZCNT. Some older software continues to use the "ffs" gcc intrinsic, which corresponds to TZCNT, even if it is trivial to rewrite it to use LZCNT instead. In old gcc, "ffs" was available instead of the more useful "LZCNT", because the ancient DEC VAX computers included a "FFS" (find first set bit, starting from the LSB) instruction and the gcc "ffs" mapped directly to the DEC VAX instruction.
Haswell (2013) has added both LZCNT and TZCNT, being the first Intel CPU which supports them.
Unfortunately, even if Haswell is more than a decade old, the older Intel Atom CPUs until Tremont did not provide support for these instructions, so they have become ubiquitous only since 2021, even if all non-Atom CPUs have supported them for more than a decade (up to 18 years for AMD).
There have been manufacturing 'incidents' in the past, also with audio CD's. Some of my old Aphex Twin CDs rot because one big UK manufacturer screwed up production in the early 90s.
Belsat's article[1] is a bit more detailed and I found this fragment interesting:
>According to Andrei Kramarenko, a leading expert at the Institute of Transport Economics of the Higher School of Economics, passenger traffic in Russia has almost stopped growing due to insufficient supply. In September, Russia's Ministry of Transport revised its annual forecast to last year's level of 103 million passengers carried by aircraft.
>Seat occupancy on Russian passenger planes this summer reached 93-95 percent. The expert believes that the situation will fail to improve next year, which may prompt airlines to simultaneously reduce average annual flight times and maximize redistribution of their share of seat supply toward the summer when effective demand and ticket margins are several times higher.
So basically it seems that their airlines are stretched thin.
The comment you replied to says that they can’t meet demand due to insufficient supply (due to grounded aircraft). Very high occupancy is consistent with that. If Delta suddenly had half as many aircraft they would always be full too.
In theory yes, in practice deliveries of domestic aircraft have stalled. The Sukhoi Superjet 100 for instance hasn't had a delivery in 2 years and I think the Tu-204 also did not get any deliveries though they expected some this year. It is however also very hard to find out what exactly is going on, because the information regarding civil aviation is quite restricted nowadays.
They have a big problem with producing modern engines and won’t catch up with the demand without Western suppliers for at least a decade, which means that in a couple of years this is going to be a huge problem.
But what AI brings to the table is how much easier it makes it to combine and analyse all that data, build peoples' profiles. It scales infinitely better than hiring human analysts to track someone. It's too cheap and too tempting.
reply