It's a vanity item, like a Rolex watch, and its existence is not for actual utility.
Utility-wise for the cost, it's not outstanding compared to regular calculators. If you spend enough time with S100X for the cost to be justified, then you are wealthy or you are spending way too much time at the calculator and should reconsider your workflow, e.g. using a spreadsheet or Python script instead, and those things are cheaper than this calculator.
Nothing wrong if anyone wants to buy it. But technically there is nothing special here, just the physical appearance/build.
This article is only saying that ZFS can mitigate disk data corruption caused by bad RAM, mainly through using checksums, not that it can completely prevent disk data corruption.
Also, it does not talk about the scenario where the in-RAM data being corrupted does not come with checksum. For example, data received from the network by the NFS/SMB server to be written to a file, before it gets passed to ZFS. This data is stored somewhere in RAM by the NFS/SMB server without any checksum before it gets passed on to ZFS. ZFS does not do any work here to detect or repair the corruption.
So, ZFS does not prevent on-disk data corruption caused by bad RAM, and only mitigates it. Using ECC RAM results in a huge relative reduction of such corruption, even though some people may consider the non-ECC probability to be already low enough.
Hanlon's Razor is such an overused meme/trope that it's become meaningless.
It's a fallacy to assume that malice is never a form of stupidity/folly. An evil person fails to understand what is truly good because of some kind of folly, e.g. refusing to internally acknowledge the evil consequences of evil actions. There is no clean evil-vs-stupid dichotomy. E.g. is a drunk driver who kill someone with drunk driving stupid or evil? The dangers of drunk driving are well-known, so what about both?
Additionally, we are talking about a system/organization, not a person with a unified will/agenda. There could indeed be an evil person in an organization that wants the organization to do stupid things (not backup properly) in order to be able to hide his misdeeds.
It's fair that Meta does not have to pay the legacy big media racket the protection fee, but unfortunately this hurts the players that were not asking for the protection fee in the first place. Maybe small players will hurt from the reduced exposure, to the advantage of the big players.
Maybe they can develop their own high-tech fabrication technology to catch up, but I wonder how they will manage and accelerate the progress.
The West has IP laws to incentivise R&D by the free market. In China, IP law enforcement is hit-and-miss. Small entities risk having their research getting used by others for free so it would be up to big entities that are well-connected to the centers of power to do the R&D. Research done by a few big entities in a centralized manner may not foster much innovation.
So maybe it boils down to how well they can spy or poach talent or knowledgeable people from Western companies, or maybe reverse-engineer Western products.
But there is no database keeping score reliably on all the nefarious actions. If a lot of companies already behave this way, then the behavior becomes normalized. None of these companies get ostracized. At worse, they just make a better offer to future candidates to attract them. In the current capitalism, everything has a price, even honor.
That's a rather strong assumption about other state actors, in terms of their current progress, future progress and their ability to attract researchers who get frustrated by the pause.
After the 6-month pause is over, then what? Another 6 months? What is to be hoped to be achieved in those initial 6 months other than non-participating entities getting a leg up? If there is no guarantee that the 6-month pause will not become one year or a few years, then maybe some researchers will move to non-participating organizations/states in anticipation of an extended pause.
From a quick reading of the wiki, the associated methodology seems rather limited:
"System complexity, particularly in software systems, making SIL estimation difficult to impossible"
"The requirements of these schemes can be met either by establishing a rigorous development process, or by establishing that the device has sufficient operating history to argue that it has been proven in use."
You could prove that normal code satisfies some specs, but you can't do that with neural nets unless the number of possible inputs is tiny. So, the only way to establish that the black box neural net meets some SIL target is through "sufficient operating history".
And someone will decide that a 0.0001% probability of an airplane crash caused by a bug is acceptable? Maybe 0.0000001% is more reasonable? In any case, how do you accurately determine the probability of a rare failure of a black box without doing many real experiments with real inputs?
The "concrete examples" he gave in the "Ongoing transition" section are traditional AI tasks. What about other tasks like program compilation and theorem proving that require more hard-core logic? I will be more convinced if neural networks can reliably do those tasks. If not, at best a human will still need to manually break the high-level task into smaller subtasks to be solved by neural networks and then somehow glue the parts together.
For the "Reorder floats + Delta", I wonder if he did anything special to prevent floating-point errors. Floating-point operations (e.g. minus followed by plus) do not necessarily round-trip because their results are not exactly mathematically correct, i.e. sometimes y + (x - y) != x.
Only if their job depended on it. For some people, their jobs mostly depend on following orders, official policy & "best-practices". The job reward is not directly related to the organization's or society's desired outcome. Codified rules and "best-practices" cannot cover everything important. Wisdom & truth are ineffable but sometimes people pretend that everything can be codified, but if that were the case, everything can be done with robots and computers.
What are some of the intricacies in the 45-cent chip that makes it hard to produce by a large number of different fabs?
In the absence of more details, it seems to me that a cheap, 45-cent, chip should not be so complicated that only a small number of companies can produce it on-demand.
In this modified version where the for-loops get converted to do-loops (possibly unsafe/incorrect behavior), unnecessary multiple vpbroadcastd's for ymm1 are still done:
With the original, for-loop version, it could be argued that in case no loop iteration gets run at all, the vpbroadcastd's can be totally skipped, and to generate extra code for different cases of whether each loop is empty to avoid unnecessary vpbroadcastd's is not worth it (e.g. the greater icache pressure resulting from longer code).
With the do-loop variant, both loops will get at least one iteration so the compiler really could just unconditionally do the vpbroadcastd once before the first loop. It somehow fails to realize that ymm1 did not get clobbered between the two vpbroadcastd's.
Utility-wise for the cost, it's not outstanding compared to regular calculators. If you spend enough time with S100X for the cost to be justified, then you are wealthy or you are spending way too much time at the calculator and should reconsider your workflow, e.g. using a spreadsheet or Python script instead, and those things are cheaper than this calculator.
Nothing wrong if anyone wants to buy it. But technically there is nothing special here, just the physical appearance/build.