Mithriil

Did you just

The visualization helped me realize something: The index doesn't change after the swap, so you always end up with the biggest number "in hand" after a full deeper level iteration loop.

This looks like a cool progress bar.

Being read by one human: you deepened their understanding of the thing.

Being read by one OpenAI crawler: you deepened the understanding of ChatGPT, thus offering knowledge of the thing to everyone that interacts with it about the thing.

One of these scenarios has much more collective impact than the other.

It is helping out the collective to give AI a tutorial on undocumented matter. Yes, you don't receive anything in exchange (except personal achievment and growth from writing, as many have commented). But making a world a better place is more important for some people than personal fame and recognition.

If AI systems are crawling everything we write publicly, we have a certain responsibility to teach it the right things. It's like our common child we're raising through what we leave open on the internet.

Good questions were asked.

Also, the discrepancies with other areas of research should be worrying:

> None of the 331 authors disclosed COIs [(Conflicts of Interest)]. Not a single one.

> This is surprising because:

> COI disclosures in other areas of research, such as bioscience, range from about 17% to 33%;

The second part of the title is more explicit: Compilation of Julia code for deployment in model-based engineering

Quantum computing is a generalization of classical computing. Thus, they CAN do classical computing. But, in practice, it'll be not as fast, more error prone and at a bigger cost.

How the H devices (which they call tetrons) form a qubit is explained more thoroughly in their ArXiv article: https://arxiv.org/abs/2502.12252

It doesn't detect a single electron, it detects parity of the number of electrons.

I'm reading it while waiting for the compiler.

> "a senior administration official tells Axios."

Another source of misinformation, now, maybe?

Whole model is 671B parameters. Downloadable from Huggingface, with 163 LFS file of around 4.3GB. Around ~700GB total.

Recommended RAM: more than most PC.

This is close to an idea I had in the past.

I bought a keyboard last year that has a small embedded CPU that I can reprogram with QMK [1]. I designed an objective/score function for key placement, based on frequency of letters (and groups of 2 and 3 letters) and finger movement (in the case of strings of letters). Each finger was weighted by how much I didn't want to move it haha. I used this function to optimize my "perfect" keyboard and reprogramed the keyboard to match it. I'm still sometimes trying to learn it to this day, I don't use it often enough. My biggest problem with it is that it is an ortholinear keyboard... Not confortable enough. Anyways.

I had plans in the future to make a small keylogger on my PC to follow my ACTUAL usage of keys, and run the optimization again automatically. Then the software could message me what would be changed and upload the layout on the keyboard every X month (not too often, y'know).

I'm waiting to get another keyboard for this, probably a split keyboard.

[1] https://docs.qmk.fm/

Energy wasted to secure the chain

To follow up on the idea.

It HAS a physical tie: it is the energy spent to run the chain, which is, however, backward to gold being a material. And it's not to the advantage of the owners, but of the miners.

I can easily see a catastrophic scenario out of this: a downward spiral of value happens for XY reason, and then a halving happens. That makes the mining of it suddenly highly costly, which makes it uninteresting, or less and less profitable to mine, which renders the chain slower and less popular, etc.

However, I think that the popularity of Bitcoin is still on the rise, and we won't see such a downward spiral before many halvings (20 years? That's a lot of GWh lost to useless computing...)

BTC is not sustainable, that's true. However, the world is now aware of decentralized ledgering tools, and other projects have worthwhile technologies and use cases.

I just wish they would drop the energy consuming ones...

I don't believe it is zero-sum.. Or at least, not rigourously. I feel the Market Capitalization of BTC must have been bootstrapped at some extent. It seems impossible to me that the sum total of what people have paid to have their BTCs to be more than $2.02T. In other words, the value the market gives it has given value to BTC out of nowhere.

A part of the valuation comes from miners who gets their returns cut by half every ~4 years. They have an interest in having the price (at least) double by then.

Interesting arguments against the premise of the No Free Lunch theorems: our datasets are not uniformly distributed across the space of datasets. Hence why some models seem so applicable to many problems, such as transformers.

Moore Curve [1] is a sequence of curves that ends up filling a square as the sequence goes to infinity, thus asymptotically mapping [0,1] to [0,1]^2. (phrasing might not be mathematically correct)

So, my intuition tells me that what the site creator did is take the 4th Moore curve (which has 256 "corners") in the sequence, then spread 256 points on the [0,1] interval and map the later into the former. The output points then serve as cores for the Voronoi diagram [2].

"Adding holes" in the 256 input points thus means to creates gaps in the 256 input points, possibly at regular interval. The animation happens when points slide from adjacent from the gap into the gap. I can't find the exact logic of which point moves when. However, all points move at the same time when there are exactly 128 holes (i.e. 128 points on the curve), and there are three batches of points that move one after the others when there are 64 holes (thus 3 batches of 64 points moving).

[1] https://en.wikipedia.org/wiki/Moore_curve

[2] https://en.wikipedia.org/wiki/Voronoi_diagram

The graph in the first figure shows negative Kelvin.. Can someone explain how that is possible?

In the conclusion, I like the image:

> My mental model [for GPU threads] is that you’ve got a bunch of container ships that can travel at 10% of the speed of light. You’re using them to ship goods around the world. They’re very fast so most of the work is in setting up your harbors so that you can load and unload these container-ships in fractions of a second so that it can sail to do the next thing. It’s not easy to feed these beasts, but if you do it right you can do huge chunks of work in almost no time.

It's not a full solution, but meditate. When everything seems too much on the outside, we can find space inside. It makes it easier to deal with the job afterwards.

A small sound effect I used once. At my last job, I queued a beep in my terminal at the end of the company-needed opening script that took 3 minutes so that I could look away and know when it is ready. (Useless to say that I almost never closed the console.)

Best we can do then is keep ourselves up to date and give our support!

The answer is: it is not aimed at Earth.

When such solar flares happen, how long does it take for the impact to be felt on earth?

EM waves is probably around 8 minutes (number I heard for visible light), but then the solar wind goes slower, I think.

The Three Body Problem (Liu Cixin) for my best friend. It's some of the best sci-fi I read.

The Courange to Be Disliked (Ichiro Kishimi, Fumitake Koga) to a friend that was struggling with social anxiety.

Consider looking into https://80000hours.org/.