define AGI
31 karma · joined September 3, 2025
the reason slug has two bands is precisely because of its anti-aliasing. you having only one implies you do the AA differently, or are less efficient with searches off the main direction.
for readers: the slug shader traces two rays, one horizontally, and one vertically, to find out if a pixel is inside a glyph or outside. one would be enough for pure inside outside, but for anti-aliasing purposes it's also helpful to know how far away you are from the closest edge. but if you have a horizontal ray running parallel to a horizontal glyph edge (not uncommon), the ray glyph intersection will return no value at all. so slug traces two rays and blends between them for anti-aliasing that always works in both cases. the bands mentioned here are just acceleration structures, basically a list of cells where each tells you which parts of a glyph are contained within it.
it's still approximate. a true ground truth would just supersample and do many point-in-glyph tests within one real pixel - at edges some of them would be inside, and some of them would be outside, giving you a smooth value to display depending on the shape of the glyph within the edge pixel.
a suboptimal trait evolving that still provided an overall population benefit (via survival or reproduction) with further improvement close to impossible without a multitude of large changes is literally just a local maximum in that high dimensional space that maps traits to whatever success metric you're measuring.
the infinite wall you describe is not a wall in that space. actually I originally wrote its more like a wall in the derivative space but its not quite right either, those just point straight down from the maximum.
anaerobic lactic and alactic regime training just has a hard metabolic and fatigue thresholds where you won't see any improvements if you add volume.
aerobic regime training (which is by its nature low intensity) does, from what we know, not really have the same limitations. on the contrary a lot of its most important adaptions keep increasing with absdurd training volumes and only show up after accumulated years and decades of training (long distance endurance athletes can often peak or maintain near peak shape into their 40s or 50s because of this).
all of this necessitates large volumes of low intensity training.
those apps tend to be vastly more architecturally complex in almost every way compared to your average corporate or web app too.
the llm must be able to output the correct code when the input says something like 'place object x to the left of object y' vs when it says 'to the right'. there is an absurd combinatorial space of the possible inputs vs outputs it must generate -> the whole set of these, interpreted from a human point of view, you could call knowledge. the [x] in input x output you could call the relationships. And the fact that the LLM doesn't have to brute force represent all of them (impossible in the limited embedding vector x internal representation state) you could interpret as 'understanding'. But again, these are loaded words that are somewhat meaningless when looking at what an LLM does in a literal way.
They have plenty graphics code to train from, so this structure will be directly or indirectly available in the training data in a very plentiful way.
In literal llm transformer terms, the embeddings must have some of their components statistically represent spatial structure in some way that later in the internal layers of the LLM give some statistics of how likely it is to occur for certain code tokens to appear relative to the input of spatial wording in the token stream.
it is likely that their internal representations encode something more general than specific input x output stream combinations (because we already know this is the case for regular words and concepts). if not in the embeddings, then a couple layers into the network for sure.
they're paying for the books, no shady things going on there. whether the publishers should deserve more than a single copy's worth is a separate question.
having the law such that its illegal to scan a book and then keep it, but legal to scan it and destroy it - gg no re there, law people retardmaxxed themselves as they tend to do with anything related to digital data.
The same "person" that shows absurd behaviour like that, assuming it's not just rage/clickbait, has absurd behaviour in a completely technology disconnected world as well.
The only thing that might differ is that it's not as outwardly obvious. Those behaviours are a result of who they are at a person (whether this be because of genetics and environment, poor soul RNG, or something else).
To that extent all technology does is make it more obvious. Basically the average NPC has more opportunities to reveal themselves as such. It's not indicative of a fundamental problem.
There is giga hyper demand for various computing products and this will likely continue for many years to come - is it likely that we will have some serious competitors to the big manufacturers (for evertyhing from silicon to the end products) a couple of years from now? Or is that unlikely based on the sheer expertise, capital, and time needed to get something out the door?
- get understanding of ordinary vector linear algebra.
- understand what vector dot product does and why
- understand why an orthogonal set of basis vectors for the space you're working in is useful / what properties it has / how its used. like basic euclidean 3d space (1,0,0) (0,1,0) (0, 0, 1) basis vectors.
- get a refresher on basic calculus, in particular integrals
- understand this inner product, it's a generalization of dot product, except you can think of your vectors having infinite number of dimensions now.
- the properties of the dot product you know (like that two vectors are perpendicular if their dot product is 0) work for the inner product too. or perhaps its better to say that the general inner product is defined to have similar properties
- there are functions that are orthogonal to each other in the same way vectors can be orthogonal to each other, and you can use the inner product to tell which ones.
- spherical harmonics are constructed / by design orthogonal to each other. how to show this and where the intuition for finding them could come from is a whole topic...
- but once you have it, just like you can project vectors onto basis vectors (to essentially transform them into the coordinate system described by those basis vectors), you can project functions into the coordinate system represented by those orthogonal functions.
- then you have to figure out why you would even want to do this. in short is has a lot of useful properties/applications. in the graphics case you can compress some quite complex functions into just a few coefficients using this (not perfectly, there is some 'information loss', but still). integrating over two functions becomes cheaper when they are projected to SH basis. it lets you do some unintuitive stuff like combine light that goes into different directions into one common set of coefficients.
for example, predicting a different branch can lead to different memory access patterns. maybe you can get rid of most of the real world cost of the misprediction by just forcefully prefetching the memory the send() path is going to use.
perhaps it's easiest to think about regular image processing because it uses the same hardware. you can think about each pixel as a particle.
a typical 4k (3840 x 2160 at 16:9) image contains about 8 million pixels. a trivial compute shader that just writes 4 bytes per pixel of some trivial value (e.g. the compute shader thread ids) will take you anywhere from roughly speaking 0.05ms - 0.5ms on modern-ish GPUs. this is a wide spread to represent a wide hardware spread. on current high end GPUs you will be very close to the 0.05ms, or maybe even a bit faster.
but real world programs like video games do a whole lot more than just write a trivial value. they read and write a lot more data (usually there are many passes - so it's not done just once - in the end maybe a few hundred bytes per pixel), and usually run many thousands of instructions per pixel. I work on a video game everyone's probably heard about and the one of the main material shaders is too large to fit into my work GPUs instruction cache (of 32kb) to give you an idea how many instructions are in there (not all executed of course - some branching involved).
and you can still easily do this all at 100+ frames per second on high end GPUs.
so you can in principle simulate a lot of particles. of course, the algorithm scaling matters. most of rendering is somewhere in O(n). anything involving physics will probably involve some kind of interaction between objects which immediately implies O(n log n) at the very least but usually more.
consider that subjective experience - to put it in the weakest and most general statement - clearly has a physical component. I'm being careful to not say that it is a physical phenomenon, is caused by physical phenomena, and so on, because while I think that's a reasonable assumption, we technically have no evidence for it.
but we do have plenty of evidence that, even if it is some supernaturally created magical process, subjective experience interacts with the physical world. for one, it clearly exchanges information with basic physical systems in your body - if it did not have some way to exchange information about what your eyes are seeing, you wouldn't be able to experience sight.
subjective experience is also easily altered with simple physical phenomena like chemical substances in your brain. so either these physics directly modify your subjective experience, or the subjective experience you have is mostly a physical product of your brain and the subjective experience part is only the end point of the process that receives all the information.
it's interesting because in physics, any exchange of information implies the existence of some directly measurable physical process. anything that is the product of such a process, you can generally speaking measure. all the things you can measure in an experiment are the things we eventually call the fundamental components of nature - like the charge, spin and so on of particles, as well as their place in time and space.
so subjective experience is either already some part we haven't observed of those fundamental components - which would in some way imply that everything is subjectively experiencing all the time - or it's an extra element we have not yet observed, but may be able to directly observe in experiment in the future.
you don't need to go to quantum physics or general relativity for everything, that would mean that a description from first principles is impossible for any field that doesn't (yet) directly connect there like biology, psychology, you name it. which may be technically correct but clearly not in the spirit of the idea.
in general this article has a quite surface level focus and narrow scope. there isn't even a description how the most common physical rendering primitive ever, a ray of light, is related to electromagnetic waves in mathematical terms. the article is not very mathematical in general which is disappointing.