There may be an intermediate range in between the ends, but it isn't clear how laundering a bunch of code through a neural network avoids creating derivative works. I'm not saying it is an impossible problem, but something that sometimes blindly spews out copies of code it has seen, complete with matching comments, almost certainly doesn't achieve that bar.

The only Microcenter in the SF Bay area closed nine years ago. The closest Microcenter is a 6+ hour drive away.

With Fry's gone too, just about the only in person computer parts stores that remain are Central Computers and Best Buy. Central Computers is pretty good, if a little pricey,and Best Buy is more competitive than it used to be, but neither is as good as Microcenter.

The severe parts shortages seem to finally be clearing up, but in the past year the issue has been GPU availability. Usually for anything beyond the low end you're better off building rather than buying from a value standpoint, but the premium you'd have to pay to actually get a modern GPU standalone has reversed that.

Two huge differences between water and electricity:
1. Electricity storage is much harder than water storage, relative to daily grid demand. Grid supply needs to match demand ~instantaneously, whereas reservoirs can balance supply and demand across months or years. Yes you can do various forms of power storage, but those tend to be effective on the scale of a day or two rather than months -- they're decent at shifting solar supply from 3pm to 7pm, but not from March to July.
2. Electricity production is limited by infrastructure, more so than water production. For the former, you can build another power plant. For the latter, past a point you're looking at approaches like desalinization which are extraordinarily expensive compared to baseline.

Building more power plants is a really effective way to get more peak electricity on the grid, but for such plants to be cost effective there needs to be demand for the power that they generate. Paying demand that exists at all times except for the 1-2% worst peaks is just about ideal for investments in baseload generation and infrastructure.

The big issues with crypto mining are firstly that it is ultimately a waste (and so more atmospheric carbon, environmental impacts, etc) -- it's good for the grid, not for the environment. Secondly the demand isn't that reliable because crypto prices vary enough from year to year that it isn't clear that it will be economical to power miners at grid rates a year from now.

It sounds like you're talking about Communications Decency Act Section 230.

Section 230 simply does not work that way. This platform/publisher business around moderation (as a matter of US law, and section 230 in particular) is a pile of lies propagated predominantly (but not entirely) by alt-right politicians and media.

Even without Section 230 the operators of these services would almost certainly still be protected in almost arbitrary moderation of their systems by the first amendment, and the main effect is to avoid a bunch of legal fights over exactly where the boundaries to that are.

You must be new here. Welcome to Slashdot!

Trump was, and continues to be, a habitual liar, in a way that no president in modern history was.

The left would have taken him more seriously if he didn't tell blatant obvious lies and falsehoods on a daily basis. These were lies in direct contradiction of the memory of anyone who can remember even a few years into the past. Practically the first thing he did at the beginning of his presidency was to lie about the size of the inauguration crowd relative to that of Obama.

consider: https://www.nytimes.com/intera... from the first year, https://www.washingtonpost.com...

Normal presidents lie on occasion. It's part of the job. Part of the trick is that if you want to be believable when you tell the lies (or even truths) that actually matter you need those lies to be rare.

TCP already has your 20:1 style problem even without Google doing anything special. Cubic, along with most other congestion control algorithms, provide fairness in the form of competing flows having equal sized windows. That means that a flow with 1/20 the RTT across a bottleneck will get 20x the throughput.

Who has 1/20 the RTT? The companies, like Google, serving lots of traffic behind large CDNs, compared to someone just trying to download from a faraway server.

Meanwhile the head of line blocking issues that HTTP/2 and QUIC target? The usual workaround for them is to have multiple connections, and multiple connections means that you have N flows that in combination compete at N times the weight of a single connection.

The largest issues here involve head of line blocking, both for QUIC and for HTTP/2.

Yes, TCP does a great job of achieving line rate, when properly tuned. If all you want to do is download/upload a large file, TCP will saturate your 10 gigabit network connection just fine once you replace congestion control with something a little bit less loss sensitive like BBR.

The problem is that for usecases like downloading a web page, you have two head of line blocking issues:
1. In HTTP/1.1, pipelining lets you have multiple requests on a single connection, but the server has to respond to them in order. That means a server can't respond to the second or third request promptly if the first is large or takes a long time to process at the server. HTTP/2 fixes this. That ultimately makes things easier for both web clients and servers that care about performance, given a robust HTTP/2 implementation. The alternative is to have a bunch of parallel connections, which is both more resource intensive and is less friendly with respect to congestion control, and unless you never pipeline at all you can still get unlucky.
2. TCP being a single stream is more sensitive to single packet drops, which is inefficient when you logically have several streams. If your higher level abstraction is multiple independent streams (as with HTTP/2) switching from TCP to a multi-stream transport protocol like QUIC (or e.g. SCTP) packet less has less latency impact because it only affects the streams where the packet loss occurs rather than blocking all streams.

Note that neither of these makes a difference if all you're trying to do is download a single large file. They *do* make a difference if you're downloading a 2 GB file and then decide, on the same connection, that you want to fetch a 1 KB piece of metadata.

Lastly, I don't think the centralization argument holds a lot of water. There are plenty of sources of complexity in implementing a web browser or web server. If protocol complexity is significant, presumably you use one of a number of existing well tested implementations rather than rolling it yourself.

As a hint: you're reading the words of a tech reporter talking about a conversation with someone focused on this topic. Assuming that "Google" is being stupid because of the words of a tech reporter, instead of the tech reporter either misunderstanding or miscommunicating, is a little odd.

There's a clear interpretation: Passwords suck. They are broken and insecure as used in practice. Sure, most of the alternate factors have their own downsides too, with SMS based OTP being notably even weaker against targeted attacks than passwords, but it would be nice to move to a world where you're still multifactor but the password isn't one of those factors. Heck, even single factor based on prompt on a phone with a secure enclave chip would be almost strictly better than the current password centric world as long as you have a story for avoiding single points of failure.

A fairly big one is having the convenience of a single filesystem spanning N disks without having everything fail and needing to restore from backup whenever a single disk goes.

If you have e.g. 6 disks (for example, bulk media usecases) either you're stuck with managing 6 different filesystems or you have something that blows up if any of the 6 disks fail, so you're dealing with failures 6 times as often. (and if you're going to do that, you might as well at least RAID-0 for the performance gain!) RAID-5 gets you something that looks, for the most part, like a 5x larger disk that doesn't explode whenever a single disk goes bad.

How is Roblox "the early internet"? 2004 is several years newer than than my slashdot account, and 35 years after the first RFCs.

https://www.census.gov/quickfa...

White 76.3%, White not Hispanic/Latino 60.1%, Hispanic/Latino 18.5%.

While the time is coming that non-Hispanic/Latino white people won't be a majority, we are still generations away from either white people not being a majority or non-Hispanic/Latino white people not being a plurality.

To what extent are we not seeing substantial fomite transmission because most people are washing and sanitizing their hands more?

I don't see the duh? There's a lot of counterbalancing non-obvious to this, especially for what was then a very new disease.

The big two things we have a lot more data on now than we did then are:
1. They were worried that telling people to wear masks, especially people who weren't used to doing so, would cause them to touch their face more and lose the protective benefit. It's legitimately possible that would accelerate the spread of the disease during the critical early weeks. With further data it turns out that fomites don't seem to be a major source of spread, so that's not so bad. Of course, some of that may be better hand hygiene.
2. It turns out that transmission really actually is driven by presymptomatic/asymptomatic carriers, which means that source control for the general population is more important than they initially thought.

(and now #3 is that there's data coming in that, while the effect of wearing a nonmedical mask in keeping you from being infected entirely is fairly moderate, they very likely reduce initial viral dose and result in a more mild case.)

We've got enough data that this is extremely clear cut *now*, but it wasn't at all clear cut in February and March, back when the dire PPE shortages *were* clear cut.

Almost the entire first world has this disease under control, on a trajectory to bring things down to zero or at least sustain very low transmission rates.

Somehow the US as a whole is on a trend upwards, and has been terrible at best at managing it despite being one of the richest countries in the world. Its national budget is larger than the economies of most countries. Meanwhile there's been fuckall national response to this thing. Closing the barn doors when the disease is already here. Weeks before the China ban there was already community spread within miles of my home.

But that's months ago. This isn't even a matter of "trust" or "trustworthiness" -- the lowest bar in the world here is to learn even the most basic lessons from the rest of the world.

The most basic intervention here? Masks. We absolutely now know that masks are effective at reducing spread from people who are infected. They are very likely effective to the point of making this thing go away like magic if everybody who reasonably can just wore them whenever possible. The thing is, they don't even *need* to be that effective. Even a 20% reduction in R_0 would allow for current levels of reopen while maintaining a downward trajectory in case rate.

Except, for some reason, mask wearing is a political issue, and it's hardly TDS to point in the direction of Trump here. Absolutely, mask slackers would be a thing without Trump, much like they were a century ago, but it's hardly helps that the POTUS makes a deliberate point of not wearing one.