Veedrac

38 ms·

421.
by Veedrac 5y ago
Like most anti-AGI arguments, this one disproves humans, and is therefore wrong.
422.
by Veedrac 5y ago
All of these problems are due to bad regulation, not the technology. Nuclear fission was well on its way to cost-effective production back in the day. https://rootsofprogress.org/devanney-on-the-nuclear-flop
423.
by Veedrac 5y ago
This number is not credible. Are we done with ImageNet? [1] points out that the best ImageNet models that currently exist surpass the accuracy of the original labels they are trained on. Expecting to reduce error by a factor 10 on unseen d
424.
by Veedrac 5y ago
> In the absence of any possibility of hard evidence to the contrary, the default assumption is going to be that the invention would not be independently invented or productized by someone else. Obviously if the patent office ignores the
425.
by Veedrac 5y ago
For sure the actual award would be a matter of judgement, as would any challenge in court. But right now patents are judged on whether they were “obvious to the person skilled in the art, at the time the application was filed”, which is bot
426.
by Veedrac 5y ago
The virtue of the patent is that it rewards invention, and its productization, in cases where either the invention itself or its productization would not have otherwise been profitable, either because of the difficulty of inventing it, or b
427.
by Veedrac 5y ago
This is the wrong place to regulate that. If buildings need to adhere to certain fire safety requirements, you have a law that says people modifying the property need to follow those requirements. You don't make a law that says only th
428.
by Veedrac 5y ago
Ten years ago we didn't even have AlexNet; I think most people would have thought a paper like that was nuts at the time. The ten years since are what popularized backpropagation as a path to general intelligence. Who ten years ago w
429.
by Veedrac 5y ago
This sort of cynicism really grinds my gears. Graphene was first isolated and charaterized in 2004 , using Scotch tape to pull samples off bulk graphite. If you were surprised in 2011 that graphene was not already a household staple, that&
430.
by Veedrac 5y ago
This is a wild take, that this military intelligence isn't useful to opposing militaries, and that hydrogen bombs aren't a pretty damn dangerous thing to risk leaking. I'd much rather people be overcautious with respect to nu
431.
by Veedrac 5y ago
I realize this is about the least popular opinion you can express on HN, but I really strongly dislike the lack of engagement with the issue here. Online abuse, of all varieties, including sexual abuse, including of children, is not a boogi
432.
by Veedrac 5y ago
It didn't seem to matter for Graviton2, TBH. ARM servers took off the moment they got fast enough to be worth converting software. https://twitter.com/IanCutress/status/1387061860965437440
433.
by Veedrac 5y ago
*Arm Subtle but significant difference. The company is sentence case now.
434.
by Veedrac 5y ago
To be fair said phone company is the largest public company in the world by market cap, using a process provided by the eleventh largest public company in the world by market cap.
435.
by Veedrac 5y ago
*In propellant terms, for an equal payload, for a SSTO These are important qualifiers. Realistically if you tightened up the mass fraction on New Shepard, abandoned reuse, and added a second stage, the technology could probably put a small
436.
by Veedrac 5y ago
Blue Origin has spent ~$2.5B on New Glenn infrastructure, including the factory and launch structures, and likely under a billion on New Shepard. This is perhaps a bit more than SpaceX would have needed, but not by all that much.
437.
by Veedrac 5y ago
Falcon Heavy technically has a little more payload to LEO than New Glenn, but it's stuck with an extremely undersized fairing for that job, so really it's better treated as Falcon 9 that's able to go further out. For the bul
438.
by Veedrac 5y ago
Yes, SpaceX moved fast in its early years; Musk threw everything he could afford at the project and raced to the finish line, whereas Blue Origin only really stepped into high gear very recently. It's true that Blue Origin started earl
439.
by Veedrac 5y ago
People have a really weird idea of the timelines here. Blue Origin's first serious rocket was New Shepard, launched in 2015. It has been six years since then, and New Glenn, a rocket more ambitious than SLS or Vulcan Centaur or even
440.
by Veedrac 5y ago
It's true we're all working in the margins of efficiency, but that's like saying there's no difference between a Falcon 1 and Starship. Saturn V isn't close to the limits of how big you can make a rocket, nor how ef
441.
by Veedrac 5y ago
In 1957 the Soviets launched the first artificial satellite into space. In 1969 America landed the first man on the moon, using the Saturn V rocket, capable of launching 140 tons to LEO. In 2021 NASA has yet to build a single rocket better
442.
by Veedrac 5y ago
Most parts of a rocket engine are structurally pressure vessels, where this rule doesn't apply, as wall thickness also scales with size. As I understand it, the main driver of scale is that the combustion chamber has an optimal size. T
443.
by Veedrac 5y ago
But it's not incompetence, these companies have made exorbitant amounts of money doing this. Congress rewards being late, overbudget, and inefficiently structured. The companies extoll the benefits of manufacturing in practically every
444.
by Veedrac 5y ago
There's a good chance for a lot more than that to come their way. Congress has spent almost $50B on SLS+Orion for Artemis, which Moonship would make totally redundant.
445.
by Veedrac 5y ago
Falcon 9 is about half the price of competitors right now, though the internal cost is under half that again; a lack of competition has prevented SpaceX from needing to lower prices further. Starship doesn't have an unambiguous cost be
446.
by Veedrac 5y ago
Note that this is not targeted to branch prediction, but compiler optimizations. It should be done carefully, because it tends to have a larger downside than its upside, and the changes the compiler makes based on it can be unpredictable, a
447.
by Veedrac 5y ago
Branch prediction works predictably well when the branch is heavily biased towards one outcome, typically follows a repeating pattern of a short length (eg. taken once every three iterations), or is strongly correlated with another precedin
448.
by Veedrac 6y ago
Label noise will almost always give you a weaker model, and sometimes it matters whether those are merely random or due to some general confusion (eg. people mistaking crabs for lobsters). But these datasets are best thought of as research
449.
by Veedrac 6y ago
Heh, I guess this is a difference in perspectives. ML has some really shoddy datasets, like really bad, worse than you think I mean. > The macro-averaged results show that the ratio of correct samples (“C”) ranges from 24% to 87%, with a
450.
by Veedrac 6y ago
This article doesn't make any major factual errors, but it's implicating that this is some sort of surprise, and that it's a major impediment, neither or which hold up. While training on noisy datasets is an interesting probl

More ›