If it’s like Marvel sequels every year then there is a significant added training cost as the expectations get higher and higher to churn out better models every year like clockwork.
Claude gave me this unobfuscated C code for `long z,x,G;main(){for(puts("P1\n80 80"),scanf("%10lx",&G);3-z/2160;x=++z%81/8-5)putchar(5>x?!(16>>(x^-(x<1))+1&G<<5>>z/6485)^49:10);}`:
#include <stdio.h>
/
* PPM Pattern Generator
*
* This program generates a 80x80 black and white PPM image based on a
* hexadecimal input pattern. The pattern is rendered as ASCII art using
* bit manipulation to determine which pixels are on or off.
/
int main() {
long pixel_index = 0; // Current pixel being processed (0-6399 for 80x80)
long horizontal_offset; // Horizontal position relative to center
long hex_pattern; // Input hexadecimal pattern
// Output PPM header for 80x80 black and white image
// P1 = ASCII black and white format
// 80 80 = width and height
puts("P1\n80 80");
// Read hexadecimal pattern from user (up to 10 hex digits)
scanf("%10lx", &hex_pattern);
// Process each pixel in the 80x80 image
// Continue until we've processed all pixels
// The condition (3 - pixel_index/2160) continues while pixel_index < 6480
// This accounts for 80x80 = 6400 pixels plus some margin
for (; 3 - pixel_index/2160; ) {
// Move to next pixel and calculate horizontal offset from center
pixel_index++;
horizontal_offset = (pixel_index % 81) / 8 - 5;
// Determine what to output for this pixel
if (horizontal_offset < 5) {
// We're in the main pattern area (central region)
// Complex bit manipulation to determine pixel state:
// 1. Calculate bit position based on horizontal offset
int bit_pos = (horizontal_offset ^ -(horizontal_offset < 1)) + 1;
// 2. Extract relevant bits from the pattern
// hex_pattern << 5 shifts the pattern
// >> (pixel_index/648*5) selects which part of pattern to use based on vertical position
long pattern_bits = (hex_pattern << 5) >> (pixel_index/648 * 5);
// 3. Check if the specific bit is set
int bit_is_set = (16 >> bit_pos) & pattern_bits;
// 4. Invert the bit and XOR with '1' (ASCII 49) to get '0' or '1'
char pixel_char = (!bit_is_set) ^ 49;
putchar(pixel_char);
} else {
// We're outside the pattern area - output newline
putchar(10); // ASCII 10 = newline character
}
}
return 0;
}
/
* USAGE:
* Compile: gcc -o pattern_gen pattern_gen.c
* Run: ./pattern_gen
* Input: Enter a hexadecimal number (e.g., 1234ABCD)
* Output: PPM format image data that can be saved to a .pbm file
*
* EXAMPLE:
* ./pattern_gen > output.pbm
* Then input: DEADBEEF
*
* The resulting .pbm file can be viewed with image viewers that support PPM format
* or converted to other formats using tools like ImageMagick.
*/
Little nitpick. C# was created by Anders Hejlsberg who studied at DTU (Copenhagen). He also implemented Turbo Pascal. Borland was also a company founded by Danes.
In general, programming language theory is pretty strong in Denmark, with lots of other contributions.
For example, the standard graduate textbook in static program analysis (Nielson & Nielson) is also Danish. Mads Tofte made lots of contributions to Standard ML, etc.
> They aren't your 'typical' Ivy League/Oxbridge univ+techhubs.
Aarhus is an outstanding university. There are a couple of dozen universities in Europe that lack the prestige of Oxbridge but offer high quality education and perform excellent research.
Lineage? Aarhus has a strong academic tradition in areas like logic, type theory, functional programming, and object oriented languages. Many influential researchers in these fields have come through there.
I also think there's a noticeable bias toward the US in how programming language research is perceived globally. Institutions like Aarhus often don't invest heavily in marketing or self-promotion, they just focus on doing solid work. It's not necessarily better or worse, but it does make it harder for their contributions to break through the layers of global attention.
Six months of darkness is a bit hyperbolic, to say the least. The sunshine, temperature and daylight situation in Denmark is on the whole comparable to what you'd find in Germany, the UK, and Northern France.
Also, long winters? You're thinking about Canada. The daily mean temperature in Aarhus, Denmark in January (the coldest month) is 1.3 C (34.3 F). https://en.wikipedia.org/wiki/Aarhus#Climate.
By comparison, Montreal, Canada has a daily mean temperature in January of -9.2 C (15.4 F).
Yes exactly. Aarhus had Martin-Löf, Nygaard, etc. Similarly, INRIA has had many influential researchers as well as OCaml and Rocq. Talent (and exciting projects) attracts more talent. But that doesn’t mean it doesn’t exist in US. Penn, Cornell, CMU, MIT and others have had historically very strong PL faculty. My understanding is due to the nature of grants in US it doesn’t give faculty the same freedom to work on what they choose as in Europe. So you get different research focuses because of that.
AI/GenAI is unfortunately a loaded word.
Does Content-Aware Fill count as "AI"?
Does using trained data to fill content count as generative AI?
There is some nuance here that will be filled in by the courts and companies over time - just like other technologies that allowed creative mix-n-match.
I do agree that the current crop of "let's get all your YT videos and all your photographs to train our models and we will pay you peanuts by running ads" is objectively and morally wrong.
However, progress requires aligning the incentives and forcing some legal framework to compensate for training - not outright stop the generative AI train as that's simply not possible in this day and age.
Could this perform better by having the internal representation of Minecraft instead of raw pixels?
It seems rather tenuous to keep pounding on 'training via pixels' when really a game's 2D/3D output is an optical trick at best.
I understand Sergey Brin/et al had a grandiose goal for DeepMind via their Atari games challenge - but why not try alternate methods - say build/tweak games to be RL-friendly? (like MuJoCo but for games)
I don't see the pixel-based approach being as applicable to the practical real world as say when software divulges its direct, internal state to the agent instead of having to fake-render to a significantly larger buffer.
I understand Dreamer-like work is a great research area and one that will garner lots of citations for sure.
> I understand Sergey Brin/et al had a grandiose goal for DeepMind via their Atari games challenge - but why not try alternate methods - say build/tweak games to be RL-friendly?
Because the ultimate goal (real-world visual intelligence) would make that impossible. There's no way to compute the "essential representation" of reality, the photons are all there is.
There is no animal on planet earth that functions this way.
Visual cortex and plenty of other organs compress the data into useful, semantic information before feeding into a 'neural' network.
Simply from an energy and transmission perspective an animal would use up all its store to process a single frame if we were to construct such an organism based on just 'feed pixels to a giant neural network'. Things like colors, memory, objects, recognition, faces etc are all part of the equation and not some giant neural network that runs from raw photons hitting cones/rods.
So this isn't biomimicry or cellular automata - it's simply a fascination similar to self-driving cars being able to drive with a image -> {neural network} -> left/right/accelerate simplification.
Brains may operate on a compressed representation internally, but they only have access to their senses as inputs. A model that needs to create a viable compressed representation is quite different from one which is spoon fed one via some auxiliary data stream.
Also I believe the DeepMind StarCraft model used the compressed representation, but that was a while ago. So that was already kind of solved.
> simply a fascination similar to self-driving cars being able to drive with a image
Whether to use lidar is more of an engineering question of the cost/benefit of adding modalities. LiDAR has come down in price quite a bit so it’s less wise in retrospect.
Brains also have several other inputs that an RL algorithm trained from raw data (pixels/waves etc) don't have:
- Millions of years of evolution (and hence things like walking/swimming/hunting are usually not acquired characteristics even within mammals)
- Memory - and I don't mean the neural network raw weights. I mean concepts/places/things/faces and so on that is already processed and labeled and ready to go.
- Also we don't know what we don't know - how do cephalopods/us differ in 'intelligence'?
I am not trying to poo-poo the Dreamer kind of work: I am just waiting for someone to release a game that actually uses RL as part of the core logic (Sony's GT Sophy comes close).
Such a thing would be so cool and would not (necessarily) use pixels as they are too far downstream from the direct internal state!
Britain was not importing.
Britain was stealing resources from India.
It was beyond slavery.
The words plunder and pillage come to mind.
USA on the other hand has a completely different problem (perhaps opposite of what you describe - being the world’s gracious and benevolent trade partner).
If it’s like Marvel sequels every year then there is a significant added training cost as the expectations get higher and higher to churn out better models every year like clockwork.