Cornus Ammonis

@FabriceNEYRET @krishdotdev I started using Kagi on a recent project that required lots of research, worth it for the money IMO.

@krishdotdev googling a keyword with "" to force its presence. Then the 10 1st answers in facts don't have the keyword.

Name something more annoying than this.

@quasimondo @fchollet This is a rhetorical trick often used to divert responsibility for social problems from the parties most responsible, e.g. saying "if you don't want a product to be made with slave labour then why do you buy it?" when obviously the product shouldn't be made that way to begin with

@fchollet So what? If we can't tell the difference anymore and are happy with slop, we don't deserve any better. It's not like the "good old human-made content" is gone forever or out of this world - may the more interesting one win.

As slop floods the Internet and as humans start relying on generative AI more and more, it's inevitable that future models will be mostly trained on slop (except for verifiable reasoning tasks where the training will be done in sims). Culture will turn into slop remixed from slop remixed from slop

@attentionmech yes, and lots of IQ test questions have obvious cultural bias because of regional differences in how topics are taught in schools. If you want a laugh look up people asking on reddit whether you can improve your IQ result by retaking; you can, but people are in denial about it.

@attentionmech When running AI evals: how do we know the test set isn't in the training data? Is this actually testing what we think it is? Is it meaningful at all? When testing human IQ: this is unquestionably empirical truth. They nailed the testing methodology first try!

@PavelSnajdr @XorDev There's no LLM that exists currently that can competently write shaders beyond a beginner level, so you're just revealing yourself to be incompetent.

I literally don't, sorry, this is deprecated by LLMs, they already do it for me I wouldn't want to waste time on stuff LLMs can do already :D I mean I was thinking I would need to learn shaders, but that was 3y ago and since then what I needed is done & shipped remember nothing about shaders and I'm more than okay with it maybe consider pivoting :D

You need to learn shaders. Start today

@iquilezles They're also quite good at doing analysis in python, if you need that.

@iquilezles The hype around this stuff made me realize most of the big twitter dev accounts are actually webdevs writing endless boilerplate larping as engineers. o3/o4 and claude are great at that stuff. Not so much at doing anything novel.

I've been giving a serious attempt at using Cursor in a C++ code base. I might still be using it wrong, but I've only managed to get it to write code that compiles and is also actually useful, once every 20 attempts or less. When it does succeed, it's limited to very narrow tasks, never large enough to offset the time wasted by commanding and helping the AI do the work. So as of today, the more I use Cursor, the bigger the productivity loss (and frustration), very far from the advertised claims. I haven't tried other competitor products though, but I'd expect the same unless there's some model out there trained through reinforcement learning instead of basic pattern memorization? Regardless I'll keep trying though because I really want the super-powers; live is short and I have lots of ideas to try. Or is my experience an outlier, and are other C++ developers actually successful with these tools?

@kyrsive music theory first stuff can be good (it's not all Jacob Collier), but there essentially is no theory of rhythm or timbre, just notation.

it’s crazy how well the “music theory first sound design second” psyop worked, like i hope we can agree this is just wrong

@joshwhiton Sounds useful for the 30 minutes of runtime you'd get out of it before your generator dies lol.

Friend just asked me to build a locally running AI rig for him in case of a disaster. In an emergency he wants to have food & water stores, generators, and AI. Makes sense. A local AI model could talk you through water de-sal, electrical work, hunting and foraging, first aid...

@tarantulae Could this be applied to Optimal Transport?

New article: "The geometry of data: the missing metric tensor and the Stein score" (blog.christianperone.com/2024/11/the-ge…). I show how you can derive a (efficient to compute) data manifold metric tensor with the Stein score alone ! Deep connections to diffusion, score-based models and physics.

Orbit traps of fractals that can be created by facing two physically-realizable mirrors toward each other.

@pushmatrix @iquilezles Appears that it's not your internet, it's all the shaders on the page compiling. If you create an account there's an option to always show a preview thumbnail instead. Shadertoy will sometimes show an image thumbnail by default but not always, not sure what the conditions are.

@iquilezles It’s like this on all my devices. Whether connected to 5G or Ethernet

Hey, if you are the AI company/institution that has been scrapping shadertoy.com nightly for a month now, please stop now.

Running compute shaders on Android with Chrome Canary and compute.toys

@kenpex @SebAaltonen We discussed this quite a bit during dev actually. I agree that's a core appeal of shadertoy and encourages creativity, we did not give full flexibility over pipeline for that reason and tried to balance that. Ultimately, shadertoy still exists, this is filling a different niche.

@SebAaltonen But... the entire attraction of shadertoy is the fact that it is very constrained!

We finally have a compute shader playground in web. Similar to ShaderToy but with compute shaders. So many possibilities.

WebGPU shadertoy-like by me and @david_ar. We were kind of caught off guard by Chrome v113, figured it'd be another year before WebGPU release, but we've had this waiting in the wings for a while. compute.toys

@guycalledfrank First thought: compute impulse response at probe locations, either manually selected or in a grid. For a compressed respresentation, try using PCA on your probes' IRs and lerp between them. The IR comprises gain, reverb, and filter params, so no need to do something special there

Random question: how would you "bake" sound occlusion/reverb? Sound propagation is so similar to global illumination, which I have the tools to compute🤔But the problem is more similar to relightable real-time radiosity - a form factor between emitter/receiver (a lot of data)

@David_Rudnick @js_horne Photography and reproducibility in print had the effect of shifting valuation to imprimatur, provenance, (i.e. branding and chain of custody) over technical merit. Same will happen with AI "artists" since a general audience can't evaluate technical merit.

@js_horne Interesting, given almost no photography is art. An insanely tiny, statistically negligable percentage of the images generated by photography on a moment to moment basis are for artistic contemplation. Not a bad analogue for synthetic imagery's probable vector

“AI generated work is not art” in a few years will sound like “Photography is not art”. The current discourse is similar to what critics of photography were expressing throughout the 1900s. daily.jstor.org/when-photograp…

@aiekick 3d texture meshed with marching cubes and then smoothed out a lot in zbrush

@cornusammonis wpow, how did you create that ? mesh or Raymarched 3d texture ?

Same as previous, but with just the skeleton

The new Shader of the Week is 'Multiscale MIP Fluid' by cornusammonis : shadertoy.com/view/tsKXR3

It occurred to me that, after all this time, I had never done a proper take on a fluid sim, and I ought to try to do something interesting with it, no? Multiscale MIP Fluid, on @Shadertoy: shadertoy.com/view/tsKXR3