chronolitus

classic case of misaligned CI (corporate intelligence)

Oh nice! If you liked it, you might like:

yourlifeunderrome.com

The gargantuan lie that is collapsing the climate is that we told ourselves nuclear is deadly dangerous, but coal & gas are merely "dirty".

I wish these posts had more code snippets and examples, seems like a cool language though!

The design says: "Look at this nice Ferrari-branded toy car! You should have one in your garage, it would look cute next to your real car (_Real_ Ferrari, gas engine, looks like a sleek lion about to eat its prey)". In that sense, it might be perfect - for a brand that's not yet certain that it can stay true to itself in a EV-only world?

sometimes we do this thing where we pretend rail is super efficient, whereas in reality it's often more expensive than basically any other option, even in countries that do it well (e.g. Switzerland,Japan)

I love rail but it is a convenience / luxury mode in many cases, and should be complimentary to car travel, not replace it.

Honestly the level editor was just me drawing on some paper, then typing out poses and coordinates into the matrices.

But if you want to see a 4D level editor, the dev of 4D Golf made his own, love to see it: https://www.youtube.com/watch?v=5pTBSafvQ7Y

You have a great website

Thanks for the link, 4D Maze is amazing and I had somehow never seen it.

Also, here's a CPU-based prototype that led to the Hyperhell engine:

https://dugas.ch/4d_creatures/4d_camera.html

I'd definitely prefer a human made one. If you know any artists willing to make a proper version of the cover art, I'd be happy to commission one (paid, of course).

Can't wait for the hardware. Only a matter of time until someone makes the first 4D game projected to a 3D hologram.

You may need more than 1 extra axis of rotations, unfortunately. https://youtu.be/tKDMcLW9OnI?t=309

Loved this book! Part of this project started from wanting to make 4D creatures and train them to walk with RL. One interesting fact I learned is that ants would probably have 8 legs in 4D.

Why? Well, apparently ants have 6 legs because this allows tripod-gait, a simple leg movement that always keeps 3 stable points on the ground[1]

In 4D, you'd need 4 points on the ground, hence tetra-pod gait (4+4 legs).

You could of course do with less, I'd guess even as low as 1-2 if you have lots of muscles and good balance.

[1] https://dugas.ch/4d_creatures/tripod_gait.html

I really appreciate your feedback!

I tried pretty hard to increase the rendering efficiency on consumer GPUs. The biggest issue is that the main view is actually a 96x96x96 grid of "pixels" (or voxels). This makes scaling brutal: going up to 128x128x128 we'd double the total amount of pixels, to around as much as 1920x1080 resolution. Doubling the grid res to 256 would get us 16M voxels, which is about the same as two 4K displays. On top of that simple 4D object meshes scale much worse in terms of tetras than 3D objects do.

A quick solution could be to give the user a few resolution options, so they can bravely test the limits of their hardware.

So I've just modified the engine to allow you to specify a custom resolution in the URL:

https://dugas.ch/hyperhell/levels/the_bargain.html?vox_resol...

(Higher resolutions might break rendering entirely if the accel structure doesn't fit in allowed memory anymore. I was able to push it to 160x160x160 on my machines)

I'll also try to think of other ways to make the rendering more efficient, maybe a BVH instead of my simpler grid-based acceleration structure? My background is not in computer graphics, so others here might have better ideas.

10 years? These days ten years feel like an absurdly long time in the future.

I had been working on a 4-th Dimension renderer project, when yesterday I saw the hackernews post where Gemini predicts the front page 10 years from now[1], and one title was "Visualizing the 5th dimension with WebGPU 2.0".

So I figured, what the heck, might as well make the implied previous article real, and fulfill our collective destinies.

[1] https://news.ycombinator.com/item?id=46205632

"Visualizing the 5th dimension with WebGPU 2.0"

I feel so seen (my last post to hn was literally about visualizing the 4th dimension with threejs - and now working on the webGPU version)

https://dugas.ch/funderstanding/4d_camera.html

Moreso genuine curiosity than as a gotcha: A lot of comments are saying this was the wrong choice. I'd find it really interesting to hear who the nomination should have gone to instead, in your opinions.

For those like me who are curious what it looks like: https://www.designboom.com/technology/exolung-underwater-air...

Adding one datapoint here: I made several projects with Shotcut and am glad it exists, but I had to learn to expect a crash and work defensively; That's how often it happened. The most usual crash was while moving clips around, the application died and the window closed, work lost.

if true, an interesting question is - on average - how many tosses are needed for you to reach some arbitrary value (e.g. $1M)?

but, since it is somewhat like a random walk, isn't it guaranteed that if you go on for infinitely many tosses, at some point you will be very rich?

[1] https://math.stackexchange.com/a/493446

You put the finger on exactly what I find incredible about the recent progress in ML - the reason I wrote this post was to see how much I could de-mystify these state-of-the-art models for myself, and the conclusion is that (after the model is trained) it all really boils down to a couple of matrix multiplications! All the impressive results we see, they're not coming from an extremely complicated system ('complicated' like a fighter jet is, with many different subsystems, which you'd need to read many books to memorize).

Of course, there's all the secret sauce to actually getting the models to learn anything, and all the empirical progress we make to make the training more efficient (ReLUs, etc). But how many of those are fundamental, vs. simply efficiency shortcuts? And: if you'd asked me 10 years ago what I thought it would take to get the kind of output these large models are getting these days, I would not have guessed anything nearly as simple as what those models actually are.

Good point, will do!

Hi! I'm the author (Daniel). I used OneNote on some old surface tablet I had lying around, but these days I'm not sure I would use it again (for example because it doesn't support exporting parts of a page to .svg)

Back when GPT-3 came out, I wanted to understand how it works, so read the papers and made this post:

https://dugas.ch/artificial_curiosity/GPT_architecture.html

I hoped it would be simple enough for anyone who knows a bit of math / algebra to understand. But note that it doesn't go into the difference between GPT-3 and ChatGPT (which adds a RL training objective, among other things).

check out "Raised by Wolves"

This view may ultimately be right, but massively ignores the current observed trends in capabilities increase[0], scaling laws[1], and things like grokking[2]. I'm seeing an increasing amount of researchers (me included) moving to stances like: "there is a scary possibility that we may solve all the benchmarks we come up for AI... without understanding anything fundamentally deep about what intelligence is about. a bummer for those like me who are see AI as a fantastic way to unlock deeper insights on human intelligence" @Thom_Wolf [3]

[0] https://www.lesswrong.com/posts/K4urTDkBbtNuLivJx/why-i-thin...

[1] https://www.lesswrong.com/posts/6Fpvch8RR29qLEWNH/chinchilla...

[2] https://twitter.com/_akhaliq/status/1479265403142553601

[3] https://twitter.com/TacoCohen/status/1584499066410790912

To stay sane: for myself as a developer, I consider github copilot as a (much) faster google/code search work-flow. I can copy / or re-mix code I find in a google search, but it's my responsibility to figure out the copyright situation of that code.

Imagine if something like google didn't exist, and then it suddenly did. People would be saying: "This newfangled computer algorithm is giving everyone copies of my code with a misattributed licence, just by typing the function name and site:github.com !"