ekunazanu

Coolest and scariest thing I've seen in a while

The claim was for AI, not LLMs. Move 37 taught us AI is better than us at strategy games, more than a decade ago.

I think a different variant/opposite of Hanlon's razor applies when it comes to corporate or political decisions: Don't attribute to stupidity when it can be adequately explained by malice or greed.

This sounds rather obvious, but I feel people forget it far too often.

It did take a while to make, and thanks for the kind words!

Author here. Yes, as mentioned, all the colours and curves are approximations. I created it in a rush during my last summer break, to be able to submit it before the SoME4 [0] deadline.

I agree that working from RGB backwards results in a lot of back-and-forth between concepts, sometimes leaving unanswered questions. Starting with the SPD means getting to matrix transforms and chromaticity is a much longer route, but I feel is ultimately more intuitive — at least compared to most of the other ways to present it. Also, I wrote this for an audience with some (basic high school) background in math, but zero background in colours. So I'd rather it be drawn-out than underexplained. I myself didn't know much about cone fundamentals before deciding to write about it :)

[0] https://some.3b1b.co/

I agree; I don't think there's any reason to assume Jevons paradox won't apply.

> If we have to switch to something like burning the model weights into silicon to continue to make gains

I think that's already being considered semi-seriously [0][1]

[0] https://taalas.com/products/

[1] https://ir.amd.com/news-events/press-releases/detail/1296/am...

> That's a you problem. I shutdown my machine when I'm not using it.

"We designed the antennas correctly, you're holding the phone the wrong way."

> This was openai’s entire breakthrough. Making this particular model architecture larger leads to emergent capabilities

Basically, the bitter lesson: https://www.cs.utexas.edu/~eunsol/courses/data/bitter_lesson...

The difference is that water vapour is a greenhouse gas. IIRC the net warming effect of clouds is a function of altitude.

> Agreements aren’t worth the paper they’re printed on.

Pretty much. They are only as effective as the body trying to enforce it. The entire point of being a sovereign nation is nobody can force you to do anything. Now it is in a nation's self interest to not violate agreements and get along nicely, but sometimes the calculus changes and the punishment may not outweigh the benefits.

Yup, the trefoil knot is one

Depends on your definition of knowing. Sure, we know it is predicting next tokens, but do we understand why they output the things they do? I am not well versed with LLMs, but I assume even for smaller modles interpretability is a big challenge.

I usually use this: https://yqnn.github.io/svg-path-editor/

> Is a Tesla any worse than you at spotting booby trapped roads

That would've been been the case if all laws, opinions and purchasing decisions were made by everyone acting rationally. Even if self driving cars are safer than human drivers, it just takes a few crashes to damage their reputation. It has to be much, much safer than humans for mass adoption. Ideally also safer than the competition, if you're comparing specific companies.

Oh I was referring to the response of the cones itself (not the response after the opponent process). I was under the assumption there's no measured curves for the responses after having gone through the opponent process in the neuron-layer(?), so the graph and caption were conflating the response of the cones with color matching functions.

> This contains a nice introduction to biological colorspace

Looks like an interesting read, thanks for sharing!

>> yielding this spectral response: [graph with negative values]

That's what I gathered from spectral response. Usually spectral response in this context refers to the responsivity of the cones. Even when accounting for 'brain subtracting green from red' (which I assume comes from the opponent process theory) the following graph has nothing to do with it. The captions too read 'Yes, this results in red having negative sensitivity @500 nm', implying the red (L) cones have a negative sensitivity to cyans — which, again, is not really the case.

> Because there’s a lot of overlap between the red and green cones, our brain subtracts some green from red, yielding this spectral response:

No, cones do not produce a negative response. The graph shows the intensity of the primaries required to recreate the spectral colour at that wavelength. The negative implies that the primary was added to the spectral colour to match it with itself, instead of adding it with the other primaries.

https://en.wikipedia.org/wiki/CIE_1931_color_space#Color_mat...

> If Linux software starts widely adapting more dark patterns it will probably impact users across all distributions

What is preventing people from just creating a fork?

> To address this concern, the team at Google has agreed to apply their subject matter expertise to build a safe, performant, compact, and compatible JPEG-XL decoder in Rust, and integrate this decoder into Firefox.

I was not aware of this. Also judging by this and the sibling comments, it looks like the momentum didn't die despite Google's apathy. Hopefully the fact that their own team is now developing the rust port, as well as the growing support in other platforms, is enough to make Google reconsider its choices.

JPEG XL had so much going for it. Kinda sad it was killed off just like that.

How else does science get 'out of the lab'?

Everything on the site seems high quality, very much looking forward to the upcoming articles

For me, there's a certain aesthetic to 1-bit bayer-dithered images, as well as images with visibly big coloured-halftone-dots, that makes it feel both retro and modern at the same time. I want to call it neo-retro, but I feel like that term already exists.

I don't know if it's the colours or the composition (or both), but some of the pictures look absolutely gorgeous

I wonder what you think of Gabe Newell then

AFAIK Germany (and most European countries) has civil law, so court rulings probably won't have as much of an impact as it would in countries like the US

The discontentment arose because it was extremely poorly planned, and failed to address the very problems it was 'trying' to solve. It is very unlikely for a state government to implement a local currency and introduce more friction to an already painfully bureaucratic system; it is a politically disastrous move.

> Monetary policy is government policy. The ruling party is seldom uniformly popular across their entire domain.

If by monetary policy you are referring to fiscal policy, you're pretty much spot on. There's always in-fighting between states and lots of complaining about the central/federal government favoring certain states over others.

I once read that diffusion is essentially just autoregression in the frequency domain. Honestly, that comparison didn’t seem too far off.