paipa

It's Bureau Serif from Smuss Type Kiosk: https://typekiosk.smuss.studio/bureau-serif

He forgot to scrub the vendor ID field.

"And likely just creating more debt down the road"

In the most inflationary era of capabilities we've seen yet, it could be the right move. What's debt when in a matter of months you'll be able to clear it in one shot?

he said his films had been misunderstood: “My opinion is that we were doing comedies. You can laugh a lot.”

I read a few Krasznahorkai novels in past few months since his Nobel Prize, some of which Tarr has adapted to screen, and that was my biggest surprise. They were so much funnier than their reputation, my tears were rolling at times. He is such a master troll at using his slow grinding, bleak prose to deliver "unintentionally funny" moments completely intentionally.

I have never watched Tarr but now I'll have to.

I agree. It's almost all prep. In fact that's exactly what I recall him saying in an interview once, that he still gets stressed by the amount of reading he has to do before each and every episode. He never winged it and rely on his knowledge alone.

And making broad connections across topics wasn't his style anyway. He's a legend but the show can totally go on without him, and it should.

Base 6 has six digits, 0 through 5. Perfect fit for the human hand, you get place value notation and counting to 35 decimal (55 seximal) for free.

Base 6 would've been the real deal.

The puzzling thing is why they chose to adopt a term used to describe "only a scaling of power" to talk about log10 values. If 1 Bel simply meant 10x, I'd get it, but Bel has baggage and also means sqrt(10)x.

Is odds a power-like or amplitude-like quantity? If you can't tell, dB isn't the most fortunate choice. It's not like mathematicians need fake units to talk about unitless ratios and their logarithms.

The denominator isn't the issue. The context-dependent base of the logarithm is, which makes 1 Bel = 10x for some things and 1 Bel = 3.16x for others.

I've never heard of decibels used in probability theory. Did they adopt it with the same baked-in bastardizations? Please tell me +10dB(stdev) = +10dB(variance) isn't a thing.

Oh, so that's why some services have separate codes for login and transacting. Didn't know it was an EU regulation.

Still better than in-house crap.

Here's how 66% compares to the trivial predictor: "pick the team with the higher win rate and flip a coin if they're equal" gives you 64% for the 2024 season I just tried it on.

I'm not even considering home/away, let alone win margins, recent form, strength of schedule etc. I'm almost amazed this model couldn't make any use of them.

Bitcoin and every other chain will migrate over to a quantum safe fork if it becomes a practical threat.

The real value of bitcoin isn't the algorithm, it's the address balances, viewed as a more important scorecard than the rest of crypto put together. Those numbers can be moved elsewhere, if necessary.

I thought the point of bet exchanges was that you're betting against other users, so the exchange doesn't have to care how good you are or how much you're betting. Why would they add friction for high rollers? Old bookmaker habits, or they're more than just a platform provider and are participating in the markets?

I don't think starting with XYZ and color matching functions is a good idea. LMS and cone response functions are a more fundamental and intuitive description of human color response, so if you're going to bother with XYZ at all, you should arrive there from first principles, via LMS.

You only need to mix two different wavelengths to render any human perceptible color. They give you four parameters to work with (wavelength1, brightness1, wavelength2, brightness2) which makes it an underdetermined system with an infinite number of solutions for all but the pure, spectral boundary of the gamut.

I'd generate new cards such that the number of pairs on screen is anywhere between 2-5. Which means sometimes you'd create two unpaired entries (which means the turn reduces the number of on-screen pairs by 1) or two different, but paired to existing entries (turn increases pairs by 1). You could sometimes throw in generating an outright pair because why not (turn maintains the number of pairs).

Title is misleading, the intermittent fasting this paper is talking about is a brutal 350-550 calories per day restriction. That's just fasting.

He managed to get my attention starting with the cathedrals' polygons and exact bell tower positions, only to pull a "50 kilometers off, nevermind".

I get the dillemma, but I don't think it's a problem that the string of squares used to define "foward" isn't always the same as path the pawn will actually take if it chugs along.

This only happens when there's no straight path to the opponent's backrank, so you are allowed to go "sideways-foward" until you hit a square from which there is a clean forward path again.

Agreed that the past should not matter, only the current position.

Your constraint should be that if the board is the special case of regular chess, the game should play exactly the same. Allowing the pawn to move towards any side is too much because it would violate this principle.

Perhaps define forward as the direction(s) which, when extrapolated as a rook, would reach the opponent's backrank. If it doesn't exist, the direction(s) which, when extrapolated, terminate on the opponent's half. This would allow extra freedom to pawns in some funky topologies that can occur in your game, but generally follows the principle of least surprisal.

That took a corny ending compared to the "Now, as an old man I realize the only thing I can change is my underwear. On a good day" punchline that I expected.

Very cool! Looking forward to playing it.

I suggest to minimize visual clutter, because any new player's brain will be overloaded trying to figure out a board, and less is sometimes more.

I'd remove the forward direction triangles where they can be unambiguously inferred from the baseline. A lower grain contrast wood texture might be a good idea too.

You mentioned you sometimes use five tile colours but you can probably improve it: yes, the four colour theorem guarantees that :)

I wish they opened it up more and give us the tools to manage snapshots. Time Machine gives you no guarantees, it'll prune them when it sees fit, so I don't trust it. I wish I could give up my clunkier setup for APFS snapshots.

Your expense is that you've hitched yourself to an oversized loan, so you're putting money from one pocket to another, except the latter has a hole in it and you have an obligation to keep feeding it. If your cost was only mortgage interest, you'd be happier with 600k outstanding at 1% than 300k at 4% for the same house.

And even if you go to a specific place in an infinite city, the entire trip will require at most one turn, an infinitesimally rare event. No wonder we never catch them in the act.

If you want someone to not e-mail you, tell them (if that) and block them. It's that easy, unless you're baiting

No example from the ultimate, infinitely differentiable tanh(tan(x))-like family? How disappointing.

Just do it. Kagi has been my favorite $10/mo spent before this collab already.

Or use -1, 0, 1/2, 1 where the new half-weight is still a cheap bit shift.

Would be cool to see what happens if you quantize towards zero preferentially. Sparsifying the matrix should improve inference speed directly, right?