[Most Recent Entries] [Calendar View] [Friends]
Below are the 20 most recent journal entries recorded in
nancygold's LiveJournal:
| Thursday, November 14th, 2024 | |
| 11:50 pm | The real AI art Remember when you asked Dall-E mini to generate a crucifix? Dall-E mini was explicitly trained not to generate human faces. As the result it produced a deformed monster on a cross. That was the true art! As the models get better, we get less and less of the art. So I think the art comes from imperfections and constraints, like censorship. Art is the beauty of ugliness! Because perfection is banal and sterile.  Current Mood:  contemplative |
| Wednesday, November 13th, 2024 | |
| 5:53 pm | The Abortion Argument For me the abortion argument (about the embryo having qualities of a human being) doesn't exist. My argument is instead: why is it wrong to kill human beings, when they get in the way? If a person can kill the armed nigger who invaded the house to do harm, why a woman can't kill the baby which invaded her body and leeches her health, and harms her financially? Babies are parasites, since it is impossible to defend their usefulness to the host. Farmers murder vermin, boutiques burn the unsold collections. So why human babies can't be incinerated that well, when there is no place for them on the planet? Why humanity wants to create more suffering? I classify this question under computing, since I don't see fundamental difference between the humanity as a process and the processes we deal with in computing, like the memory management or killing a useless program. In this context, a birth of unwanted baby is similar to a memory leak, which evolves into memory corruption, going over the memory intended for useful tasks.  Current Mood: contemplativeCurrent Music: Like the wind/The Most Mysterious Song on the Internet (2a03 + sunsoft 5b famicom cover) |
| Thursday, October 31st, 2024 | |
| 12:17 pm | ChatGPT ![[info]](http://lj.rossia.org/img/userinfo.gif) necax claims there is a shorter algorithm to sample from a population, but ChatGPT comes with exactly the same algorithm I invented, which it calls "cumulative summing method":https://chatgpt.com/share/67236748-d It is completely equivalent to my list.outcome = $find(s~+=?0; p~^(Me{?0}.z*)<<s~).1method, except that it produces a list of K samples, instead of just one. import random def custom_choices(population, weights, k=1): cumulative_weights = [] total = 0 # Build cumulative weights for weight in weights: total += weight cumulative_weights.append(total) # Generate samples samples = [] for _ in range(k): rand = random.uniform(0, total) for item, cum_weight in zip(population, cumulative_weights): if rand < cum_weight: samples.append(item) break return samples # Test it with your distribution population = ['short', 'mid', 'long', 'bald'] weights = [70, 20, 5, 5] samples = custom_choices(population, weights, k=10) print(samples) Current Mood: contemplative |
| 12:29 am | Symta is too difficult even for me So as you know already I have to generate a lot of unique characters procedurally. And any generation involves some distribution. So I pulled out of my ass the following statistics: 70% of ladies have long hair 25% have mid hair 5% have short hair 70% of men have short hair 20% have mid hair 5% have long hair 5% are fully bald In game these are defined as a list MaleHairLength: 70,short 20,mid 5,long 5,bald These allow me to settle on the hair length of a character. I had some code to pick a random sample from such a distribution. MaleHairLength.find(s~+=?0; p~^(Me{?0}.z*)<<s~).1 That is cryptic and it does really cryptic things inside the find's lambda and (implicitly) outside of it. Now the problem is that I already forgot the implcit variables semantics I introduced years ago. Since most of the code is not command line oneliners or math shit. What is worse, I have never documented it. So now I had to go over the macros implementing these constructs deciphering them. And obviously these macros are not simple. TLDR: I need to document at least the harder parts. Current Mood: contemplativeCurrent Music: Anvil of Dawn OST |
| Wednesday, October 23rd, 2024 | |
| 1:29 pm | Multimodal agents controlling Windows 11 are available as a service https://www.youtube.com/watch?v=xr0FCUN All these office workers will have to pack up and join me at the unemployment market. The engineering breakthrough we have deserved. Current Mood:  amused |
| Tuesday, October 22nd, 2024 | |
| 4:35 pm | "А также в области балета..."  Current Mood: amused |
| Friday, October 18th, 2024 | |
| 12:27 pm | My issue with game engines When you use a framework, it dictates you the architecture and limits you. In other words, it is you who gets used as tool of the framework's creators. Game engines also invite niggers to shit out mediocrity, asset swaps and walk simulators. Since anyone can buy some FPS template and change title screen into My First Columbine Homage. Especially since making something unique requires very good grasp of the engine. So I'm surprised the niggers got angry when Godot also began dictating them political views. After all these retards can't form their own views, just like they can't author their assets. And nope, that isn't solved by forking Godot into Nazidot or switching to Unity. That is in fact a long running problem: https://en.wikipedia.org/wiki/Eternal_S The only solution lies in eugenics and genetic engineering. But wars work too, since they weed out the worst dumbfucks. Current Mood: contemplativeCurrent Music: Cosmos Elegy (コスモスの悲歌) - A mysterious Japanese waltz on 78 rpm record |
| Thursday, October 17th, 2024 | |
| 11:14 pm | The Spell of Mastery Progress Currently I have to design the skills UI. Back then I developed my own HTML clone completely with embedded widgets. Now I have forgot how it works, since I never documented it. So have to go through the code deciphering the tag meaning from their rendering. Still much easier than using HTML. For example, I can introduce table with a single tag `t [X_OFFSET_IN_SPACES]`. Compare that to HTML with its mess of table, th, td and tr. There is no way I will ever do webdev. Better being a homeless cocksucking whore. Anyway, I need some loop or dup tag to repeat an element multiple times.  Current Mood:  determinedCurrent Music: Hitoshi Sakimoto - The World For The People |
| 3:03 pm | Nigger Terms If you see an image editing software using terms "dodge" and "burn" instead of "brighten" and "darken", you immediately know it was made by the conservative niggers, who stick to their degenerate dark age technologies. Same kind of people will draw a thorn, a nail and a crucifix to represent passion, which all non-subhumans will represent by a burning heart. Current Mood: contemplativeCurrent Music: Sergei Prokofiev : The Love for Three Oranges, orchestral suite from the opera Op. 33 |
| Monday, October 7th, 2024 | |
| 11:11 am | The Main Problem of America They are still using their deficient measurement system. Why they wont abandon their medieval yards and gallons? Because traditions and patriotism and a lot of dumb niggers hostile to any change. The dumb people are the main problem of the humanity in general. Current Mood:  annoyed |
| Sunday, September 29th, 2024 | |
| 3:38 pm | Most ambitious game so far Attempts to simulate 100K population city on the Sims level with Transport Tycoon style graphics https://www.youtube.com/watch?v=Bku59Jy Last update was 6 months ago, so I guess the project is dead. Still would have loved to see a micro level transport tycoon game or some gang level GTA style strategy with these depressive commie block ghettos.  Current Mood: amused |
| Friday, September 27th, 2024 | |
| 11:59 am | Voxels on C64 Apparently 8bit systems were capable of rendering 16x16x16 voxel models, although only at a predefined angle: https://youtu.be/L8onlB0F1_A?t=310 That demo also has cute text scrolling on a ball, but I think it is done through color-map animation.  Current Mood: amused |
| 2:19 am | Spell of Mastery Progress The existing city builder UI was a bit bolted on. And it doesn't allow arranging buildings strategically. The city layout was actually generated randomly. Earlier I have already implemented monolithic structure. So why not use it to implement buildings? Well, initially I thought that will be impossible. But now I see how it can be implemented. Thanks to the rogue-like time scheduling. In fact, I can directly transition the existing city mechanics. Think Dungeon Keeper meets the Sims meets Master of Magic. The hardest part now is adapting the stock art to ingame grid. As well as cutting it into manageable pieces. Here is a shrine WIP:  Current Mood: amused |
| Thursday, September 26th, 2024 | |
| 8:44 pm | Unity I don't use Unity, but I do use a few placeholder assets from Unity store. Now Unity stored doesn't allow downloading assets without Unity installed. Unity is a huge 7 gigabyte bloated piece of shit. Although games made with Unity tend to be even bigger. Installing it alone is not enough to download assets either. One has to also create a project. Which in my case fails with "compilation errors" Would would it ever need to compile an empty template project? The Unity Editor also tends to crash even in idle, after importing assets. That is some GNU/Linux level stability. To all stock assets creators: please publish them at better stores. Like itch.io. Unity should just die already.   Current Mood: annoyed |
| 5:16 pm | The Barnes-Hut Algorithm for LLMs So currently the main issue with LLMs is that they are bound by the N^2 law, where N is the context size. The problem appears to be very similar to simulating gravity. Yet for gravity we can use the Barnes-Hut approximation, where far away entities are grouped together as a single entity. Therefore the question: is it possible to adapt Barnes-Hut to the attention matrix of LLMs? I think people are trying to do something similar (hierarchical approaches, summaries and tree of thought), but they still don't see the forest behind the trees. I.e. that the language modelling is basically a physics problem. All these N^2 connectednes/pathfinding problems appear to be expressible as gravity over some distance function, which itself a sorting problem, which can be solved by a N*log(N) algorithm. Current Mood: amused |
| Wednesday, September 25th, 2024 | |
| 11:52 am | ChatGPT finally published a truly interactive voice assitant https://youtu.be/RI4GTKMGt4s?t=132 TLDR: it reacts instantly, can speak in a chosen accent and can be interrupted. So can be used as an actual voice interface, when you're blind or say driving. And it can also be used as an actual interpreter. Current Mood: amused |
| Tuesday, September 24th, 2024 | |
| 9:55 pm | Microsoft Patented Mouse Cursor in 2007 https://patents.google.com/patent/US888 TLDR: in console RTS games (since Herzog Zwei) moving the cursor also moves the screen. Microsoft actually patented it decades after it first appeared. I think same scheme was also used in many computer RTS games, where moving cursor far enough from center made screen moving towards it. Current Mood: amused |
| 1:04 pm | People ask me why I still don't use Rust Rust is not what you think it is. Rust isn't really a programming language. Rust is a political tool, whose sole goal is taking control away from the programmers. Just like they used flu to justify authoritarianism, then use segfaults to justify Rust. So I prefer the based old school languages, like LISP, REFAL, POP-11 and C. TLDR: Rust is cringe and woke ruse, invented to enslave the programmer.  Current Mood: contemplative |
| Friday, September 20th, 2024 | |
| 1:59 pm | Decided to overload the `/` So I can use Symta as a Bash replacement. Would still need to implement pipes somehow. Using the No type (Symta's NIL) to link all manner of crap is a questionable design, but it fucking works and `no` is easy to type.  Current Mood:  accomplished |
| Wednesday, September 11th, 2024 | |
| 11:49 pm | Taxpayer Money To those god-loving christians insisting that people like me should be euthanized for being useless. Look instead how soyentists waste your stupid money: >In 2012, a group of scientists at Fermilab in the U.S. successfully demonstrated neutrino-based communication by sending a simple message ("neutrino") through 240 meters of solid rock using a particle accelerator as the transmitter and a large neutrino detector as the receiver. The message took about two hours to transmit, illustrating both the feasibility and current limitations of this technique. https://www.researchgate.net/publicatio Current Mood: amused |