numpad0
https://compoundsemiconductor.net/article/114125/Realising_red_GaN-based_microLEDs_with_europium-doping
https://pubs.aip.org/aip/jap/article/123/16/160901/400705/Perspective-Toward-efficient-GaN-based-red-light
https://www.researchgate.net/figure/Color-online-a-Schematic-of-an-InGaN-based-red-LED-epitaxial-structure-b_fig1_355092457
https://www.tytlabs.co.jp/en/english/review/rev402epdf/e402_007kato.pdfThere are innovations and discoveries made and exploited everywhere. How much things seem to be going on relative to your cognitive capacity has not much to do with the reality.
OpenAI had been miniaturizing GPT models as far back as GPT-4o era, but Chinese free-beer models that run on 16GB consumer GPUs can handle the vast majority of LLM use cases, and now they have no options but to attempt development of actual superintelligence that can cure cancer or solve the P=NP question just to differentiate against free-beer alternatives. It's clear what China is trying to achieve.
Could this be like nanostructure based? Like nanoscale etching that messes with the light, created on top of UV LEDs?
I think there might be a lesson learned from the era of attention economy and also a potential explanation as to what advertiser friendliness was about - the attention economy did not graduate from being a surplus market. Advertisers had no incentives to subsidize the waste byproducts of fried rice stall, but the system was too deeply molded for them. The amount of standalone contents that are native to social media are tiny.
Anyway, I just don't think the contents that are made because of a social feed are the ones that weren't worth much even on social media.
Most V2H systems don't support zero-interrupt automatic transition because it has no means to know if the perceived grid outage is accidental or intentional, and if intentional, trying to back it up could injure/kill electrician who commanded it. But if the system is operating off-grid, I think those can be configured to be always uninterruptible.
- JR East took the whole east half of Honshu from southernmost end of Tokyo area all the way to Aomori.
- JR Tokai aka Central got just Nagoya area + later the entirety of the Tokaido Shinkansen, Tokyo to Osaka.
- JR West got the whole Osaka area and to the west of Honshu.
- JR Hokkaido, JR Shikoku, JR Kyushu got one giant island to themselves each.
"2000 since 1990" first time I heard it
"2000 and stopped there" during 2010s
"2000 since 1980" lately, post-COVID
The problem is that all ticket systems that deal with JR Group has to be a front end to a system called the MARS System[1][2], and day-to-day operations of Japanese train and travel industry rely on its specific behaviors and intricate nuances, and that the collective of nerds and nerd-employees had adamantly refused complete changeover of the system for years. It's like you can have a webapp wrapping over the old 3270 terminal commands any way you want, but anyone with a badge and the uniform will insist that it's faster to have them hit the real analog buttons in the right order. Or just use Vi or mysql shell or something.
Some of JR ticket machines reserved for express trains with different frame colors compared to others are actually what are considered self service terminal of this MARS system rather than being ticket machines, and its UI is apparently close enough in structure that it even triggers the same unconscious habit of tapping the bottom 1/3 of the right part of the frame for retired or current employees. Which means it'll be unintuitive as it gets.
To effectively use the MARS user terminal and get JR tickets, you have to have a mental image of the ticket you are seeking to obtain, the MARS command sequence that yield that ticket, and operate the terminal accordingly, whether the terminal in front of you was the older crew terminal or newer terminal or its travel agency subvariant or the kiosk version stalls or one of numerous web interfaces or whatever it is. Some of fields such as seat numbers and choices of possible routes can be determined as you go, but in the way conforming the JR ticketing mechanisms.
I don't know enough about those stuffs myself nor I find that embarrassing. I just know enough to know it's called MARS System and if there were stations in "via" fields of a ticket that the train I will be on won't even be passing, it must be some MARS shenanigans. I certainly don't find it fair to expect someone just visiting Japan wanting to see some wooden temples and maybe try matcha, to be perfectly fluent with those. But if you were wondering why... ^this^ is why.
1: Pronounced mulls system like the Roman god. From the Wikipedia page: the original MARS 1 went online in 1960 and features were extremely limited. MARS 100 onward that replaced the original model were built on Hitachi mainframes. 305 onward were developed by a privatized JR Group company instead of the national company, JNR. Current iteration has the model number Series 500 MARS 505. https://ja.wikipedia.org/wiki/%E3%83%9E%E3%83%AB%E3%82%B9_(%...
They can probably build like, a gigantic satellite that fits in an expendable Starship that has like just couples of RTX5090 32GB backed by a Tesla battery pack and a gigantic multi panel actively refrigerating radiator system. Or use bunch of such Starships to assemble a whole rack worth of GPUs on orbit that costs billions total and minimum half a year to build out. Satellites and radiators are solved good enough for that.
Or you can put just the computer part on a desk indoor on Earth and plug it into the wall and have it cooled with fans. Or have a guy roll up an air cooled rack full of GPUs next to it. Why involve the rockets and satellites and all. What's the point. That's what everyone is puzzled with.
Radiation hardening is not a huge issue in LEO either. It is an issue, not the issue.
Overall, it's like choosing international flights as your choice of a restaurant. It's...fundamentally...doable...? But nobody knows why you would want that, unless flying GPUs is the point rather than to offer a competitive service using the server sats.
IMO, what that tells us is that a completely dumb repetition of small set of simple concepts to just set up the pathway can be an effective means to kickstart language acquisition process. Most language learning materials attempt to establish a compact set of rules to be expanded with a long tail of modular add-ons, even many that aims to create an alternative and less stressful approaches like Duolingo. The Inmu dictionary has less than 500 known expression, all unusable in regular communications and inappropriate in most settings, and the way it is used is simplistic, childish, disrespectful, and dumb. And I think that might be how "start with picture books" advice look like if rebuilt from first principles.
They can upgrade the system so that the minimum distances are smaller. Sure. Except, such hypothetical upgraded system was already in place, since 90s, and __that was what went down__.
If you were to then ask me, okay, why did they get the trains apart before committing with the downgrade into the older system. And that would be the right question. It's such a braindead move to not do that before executing the full downgrade. It's not like trains were moving when downgrade took place and it couldn't be done, the newer system going down triggered emergency stop of all trains. Train systems are intentionally built that way. So they could (load-bearing-)just ordered the trains to get spread out so there will be enough distances between trains so the older version of "train TCAS" would work correctly. The decision to NOT do that, with hindsight, is downright reckless.
By the way, if all reports on this matter is to be believed, the required distance of the older version of the "train TCAS" in operation was in excess of 25km, or 15 miles. That's long. It can be shortened. Sure. But then again, there was such a newer system already in place since 90s that has much less minimum distances than that, which went down.
LLM auto-translations haven't changed that much. The current corporate seems to be trying to figure out why and whether those remaining huge mass of foreign users lean left or right by the criteria of US politics, but still haven't have any clue about that nor had accepted that non-English-speakers actually don't speak English lol(or, to say, find translated contents natural as speeches with reasonable flows of logic, to be even more blunt)
GPS can't work for trains, they go inside tunnels and between walls. Track circuits and train sized NFC-like tags reporting linear locations, mentioned as not scaling in GP, is the standard solution. It kinda scales fine.
1: https://futurism.com/advanced-transport/waymo-crashed-into-e...
What they needed to do was to move ALL trains around until the initial state is in a valid configuration under the ASFA system. Probably like moving one train at a time at <40kph so that there is minimum one ASFA block, which might be tens of miles or however long they are, between every single trains on the line. Then the whole operation could be switched to ASFA system at full speed at much reduced capacity. It also means the train more or less wouldn't be moving for the rest of the day. Not only passengers would be angry but there will inevitably few that needs to be hospitalized for things like mild dehydration and sometimes small injuries if they needed to be evacuated.
What likely happened was that they didn't think about any of that at all, and just hot rebooted into the ASFA system without deconflicting anything. The ASFA system signaled ok to proceed to all trains sharing the same ASFA blocks to their next vacant blocks, not knowing that some of blocks were occupied by multiple trains before humans put it online, which would be an out-of-question horrifying hypothetical to the ASFA computers. And when the system ordered a train in the block to stop because a block in front was occupied, there were in fact two trains in that block, one still driving towards the one in front at 160kph, far away from the stop signal, and two collided.
On single lines with bi-directional services, you can design and implement a token based lock system: it's literally a token, often referred to as tablets, what inspired mutexes in computer science. They are like a golden medal or a durable hunk of metal, with the name of the station it belongs to engraved, often enclosed in a leather case or something. The arriving train drops the token it holds at the station, and obtains the new one when departing. The trains traveling opposite side will return them. When a train fails to obtain it, whether because it merely failed to catch it or the new one hasn't arrived, it MUST stop once, before it technically leaves the part of the rail that is considered "the station", until one is obtained.
This won't work as-is if there's going to be a T shaped crossing in which lines branch from the trunk line into short segments to stations. The crossing must be a station with its own lock token in that case. But that also can be some sort of crude technical stop for staff uses only with just maintenance guys with helmets and vests as station masters and apprentices moving the tokens around.
If they couldn't get computers handle it or make signals turn from red to blue or at least red to nothing, and didn't even have that token system, the train cannot move. It may technically be able to, and maybe it's ok for moving just few meters back and forth, but that's it. The trains should not have been operated at all if they couldn't figure out the computers.
LLMs are still text completion engines, Maybe in the way we humans are merely adaptable to spacewalks and not designed to do it, but I think wild claims like "China backdoors LLM to generate vulnerabilities for dissents" needs more substantiation than just these random experiments.