soVeryTired

Still pictures though, right? Not actual videos of someone making the full gesture twice in a row.

What would it imply about him? He’s now a quasi-governmental figure aligning himself to the far right. He’s messaging to others on the far right that the federal government will tolerate them. Doing so will embolden them to take action.

You say "wrong political party" as if he's involved in some mild political spat. The Financial Times just ran an article about how trump might defy the supreme court on his way to burning down the republic.

Musk supported AFD while tweeting about the great replacement and throwing a nazi salute. He's supporting the wrong political party in much the same way Goebbels supported the wrong party.

Or… Bob makes bread, Alice grows flowers. Alice needs bread; Bob likes flowers but can go without if he wants to. Who’s going to come off worse if they have an argument?

A lot of the field of macroeconomics, yes. Have you ever sat down and read that stuff?

But that’s under some weird, incorrect model of how money works. Presumably if we go down this route any further you’ll start talking about IS/LM and the money multiplier. All models are wrong, but there’s a better model that involves capital constraints and the Basel accords.

In any case, one of the major failings of modern macro is to assume you can think your way around the implications of economic actions. You have no way of knowing the consequences of stashing money under a mattress unless you actually do it, and even then there’s no guarantee the effect will be the same across different economies and in different eras. The world is just too complex for that.

The AI angle is weird, and a red herring. I’ll happily give 100 to 1 odds that something as abstract as topos theory isn’t useful to machine learning. As far as I know, the full weight of topos theory isn’t even needed for algebraic geometry.

The story of Grothendieck is a tragedy about a generational genius, not unlike Godel’s story. It’s deep and far reaching enough to stand on its own without AI hype making it appear more relevant.

To add to your point - you definitely can make predictions about aspects of someone’s personality by looking at their face with the expectation that you’ll do better than chance.

The paper calls out inferring political leanings as an example of pseudoscience. Give me an American’s age, gender, and race (which I can roughly identify by looking) and I’ll tell you, better than chance, whether they’re a trump supporter.

Scientific writing is bad so much of the time. The bio sciences are particular offenders in my experience.

Good writing should communicate a concept clearly. For some reason, science writers feel the need to swallow and partially regurgitate the dictionary. “Look how clever I am: therefore my conclusions are valid”.

I’ve spent ten minutes reading, and I still can’t figure out whether they’re saying that physiognomy should not be explored or does not work. They might have a good point but by christ are they asking the reader to work to dig it out.

I finished my PhD thesis with a quote [0] from Hunter S. Thompson.

[0] https://www.goodreads.com/quotes/293779-take-it-from-me-ther...

The pieces are fractal and exceptionally complex. It’s related to the idea that a point has zero length, but a line (an infinite collection of points) has nonzero length.

It also highlights a discrepancy between our physical intuition of space and the way we model space mathematically.

Snap and crackle I couldn't tell you about. But jerk is definitely important.

I used to work at a self-driving car company, and all the vehicle's motion was planned around how much jerk to apply.

Your muscles are pretty good at applying a constant force (or responding to a constant acceleration). Hold your arm out straight: it's no effort to keep your arm still and counteract the force of gravity. Now imagine gravity varies quickly and randomly between 0.5g and 2g. I guarantee your arm won't stay still.

The same prinicple applies on a bus or in a car, except this time the forces are smaller, and it's your neck keeping your head still!

There ain’t no justice.

In some situations I agree you can get a rough and ready sense of one curve or other. Certain types of auctions, and order books for financial markets are examples (though I think your link is broken!).

Even with order books though, they just provide point-in-time snapshots. It’s well-known that buyers will get a better fill than what’s shown in the book if they break their order up over time: that’s why it’s normally to trade VWAP.

These are definitely the exception rather than the rule. I work for a financial services firm and I can confidently say we have no idea what would happen to demand if we cut our fees.

Supply and demand aren’t observable though. Only their intersections, price and quantity, are observable.

Honest question: if we don’t know the shape of the curves, if we can’t observe them, and if they shift around according to unknown dynamics, what good are they?

But this is mathematics, not physics. It's about axioms and logical contradictions, not apples.

What's SETH in this context? I googled to no avail.

It's the intent to shock or humiliate behind the sending of a lewd picture that's disturbing, not the picture itself.

A urologist can look at penises all day and not care because there's no ill intent behind the display. On a dating site that's absolutely not the case.

If I sent someone a picture of a raw steak via a dating app, it wouldn't "just be a picture of a steak". There's nothing inherently wrong with pictures of steak, but the context is important.

In the context of a dating site, sending someone a picture of a steak is inappropriate and creepy. Is the user trying to communicate that they see someone as a piece of meat to be used for pleasure? Is it a threat ("I'm going to cut you up")?

Likewise, there's nothing inherently wrong with a picture of a penis. But seeing first hand the aggression and intent to humiliate or victimize that's behind a dick pic is what's disturbing. See enough of that every day and you'd lose your faith in men. That's the problem for the moderators.

I have to say, neither of those answers are very satisfying. Neither is wrong but I don't come away with a clear explanation for why cosine similarity is the obvious metric to use.

It's not obvious to me that vector size isn't relevant in semantic search. What is it about the training process for semantic search that makes that the case?

I still don't understand why cosine similarity is used (other than maybe being cheap to compute). I've read a bunch of hand-wavey articles but none of them add up for me.

If all your vectors have the same 2-norm, the cosine similarity of two vectors is a monotonic function of their euclidean distance (i.e. the minimum in one measure will be the minimum in the other).

If all your vectors don't have the same norm, where's the sense in projecting them onto the unit sphere and measuring the angle between them? You're throwing away important information about the embedding.

Could anyone explain?

Traditionally you just use money to do that.

ETFs don’t let you create the underlying asset out of thin air. They’re just a wrapper that allows the underlying to trade easily on an exchange.

The articles and courses mentioned in the WSJ piece are about science education rather than science itself. I looked up the Afrochemistry course: it's listed along side other 'fun' courses like the chemistry of art and the chemistry of cooking.

Storm in a teacup.

Everyone worked with Fortran long before they worked with Python.

PCA is purely deterministic (and might not give great results). My guess is this is done by t-sne or UMAP, both of which depend on a seed.

By that logic I could put the Iraq war on your head, assuming you're American. Hamas supporters are complicit, but not all Gazans are.

It sounds like the government might have understood it as a termite invasion.

> My theory is that architecture doesn't matter - convolutional, transformer or recurrent, as long as you can efficiently train models of the same size, what counts is the dataset

I assume you mean architecture doesn’t matter much, as long as it’s roughly sensibly designed.

The basics of architecture do matter though: a two-layer fully connected network won’t get very far no matter how many parameters it has. Architectural improvements like ResNets have definitely made a difference to the field.