superjan

I have recently been wondering how it is possible that they can make features like that, and provide closed captions, but that they continue to insert ads halfway a speaker’s sentence.

Next level: let them login and forward the ssh connection to the digital equivalent of a room full of mirrors.

The currently agreed protocol leads to < 10 transactions/sec. This can be improved, but 1000x will be challenging.

https://en.wikipedia.org/wiki/Bitcoin_scalability_problem

What I think is worth knowing is that compute units in GPU’s also use SMT, usually at a level of 7 to 10 threads per CU. This helps to hide latency.

Most touchpads can be configured to only register clicks when you actually press and get the click feedback. It should be the default. Swipes may still come through though.

In principle, any object can become a black hole if it is made dense enough. It does not matter for the gravity at long distance: replace the earth with a black hole of the same mass, and it not change the interactions with the other bodies (sun and moon, mostly).

Light does not escape in the black hole case because the black hole is smaller: in case of a black hole earth, light starting at 6370 km (current radius) can still escape. A black hole weighing the same as the earth would have a 9 millimeter radius, no light will escape from there.

Just curious but what is keeping you from compiling x*3?

Well, if we suppose all the anti-matter is in those black holes, we kill two birds with one stone. It’s ridiculous, I know, but your remark made me realize that an anti-matter black hole is essentially indistinguishable from a ‘normal’ one.

What does it for me is that turing machines can be thought of as physical. In a way it is more tangible than an electronic computer.

Btw: I think your site would look better with left justified text. Right justified looks best only for long paragraphs (books).

And (mostly) no predictions that are testable. If it is not falsifiable it is not physics.

And don’t forget cosmic inflation.

For some background: here is an interview with Daniel Lemire who built an entire career based on the observation that I/O is often not the bottlenck. https://corecursive.com/frontiers-of-performance-with-daniel...

Yes, and: “which is now predicted for around 18:56 CET on 8 March +/- 0.4 day”

Prediction specified down to the minute with but the error bars are measured in days.

Well the ship gets a little nudge in the right direction for free.

Isn’t there a normally a mechanism that lifts the wire after the landing gear has crossed it?

That’s no secret anymore, is it?

On the GPU, vectorization is done for you by the compiler, and the cores are designed to hide latency. These techniques will do little to improve that. OTOH, you have control over how the code is vectorized (stripes or tiles), that could make a difference. And it might be worth exploring if it’s worthwhile to emulate doubles with smaller types. GPU’s are slow at 64 bit math.

But, as the author already suggested, first check if there is a need to optimize at all.

They won’t. The car maker has the Budget product and the Premium product. The car maker wants to justify a high price for the Premium product, so they put in all conceivable features, useful or not. What you want is a Decent product. If they add it to the product line, it would cannibalize sales for the premium product, so they gain nothing. For the same reason my wife’s Premium electric toothbrush has a color screen, shitty IOT bluetooth features and a terrible battery life.

Bear in mind that comparing floating point NaNs is not transitive. IIRC, Rust considers floating point unsortable for that reason.

Yes, and what other vulnerabilities would be out there? IP6 is something that nearly all machines have, and have enabled, but it is rarely used.

A long time ago I practiced to use a kitchen knife properly. Rewarding AND timesaving.

If that’s what it takes to get them compliant with EU’s GDPR, shouldn’t the same apply to youtube and gmail? Or the rest of google services, for that matter?

Parser mismatch vulnerabilities.

https://www.brainonfire.net/blog/2022/04/11/what-is-parser-m...

I speculate that beyond that, extra longevity can also be used to increase your budget. If you plan for six months operations, you have a lower mission budget. They won’t shut you down after six months as long the device is still working.

The only times I worry about min/max/clamp performance is when I need to do thousands or millions of them. And in that case, I’d suggest intrinsics. You get to choose how NaN is handled, it’s branchless, and you can do multiple in parallel.

It feels backwards that you need to order your comparisons so as to generate optimal assembly.

Why is Brittney’s video not a reliable impartial account? Also note that nothing she says about her performance and feedback is in any way disputed during the call. They had nothing to say to the contrary.

It is a mixed bag, but definitely improving. Some experiences: From Holland, It was impossible to do an online booking a ticket to Italy. Only possible via phone, or as separate legs. For long journeys with changeovers watch out that you don’t get stuck halfway due to a missed connection. From NL, London, France, Germany is easily reachable, as well as Barcelona. I took a sleeper train to Stockholm last Christmas, we had delays, but made it in time. In general if you have a long distance leg, it’s best to have it at the start of your journey so you don’t miss it.

Railways do have single points of failure. A single disruption on a key piece of track can disrupt service for hours, even days. This is less of an issue with air travel, there’s usually other airlines, and alternative routes.

When there is more than one enemy, it can become worthwhile to simply use dijkstra from the player’s point of view, and then each monster can look up the optimal route to the player. It makes the computation cost more predictable when the monster count is variable.

I have seen an article posted here that did A* on a low resolution versions of the game map, and use the output of that as the heuristic for the full res version. In that design, the low res version was auto generated.

Agreed. It’s renewable, so who cares if it’s efficient?