News for hackers. Stuff that matters.

wegfawefgawefg

Re:Astra can read and write weights to improve Qwen on Tuesday November 03, 2026 @12:48AM (#50348544)
likely all it takes to make this work is for major labs to set up an rl gym where you extrude models that solve tasks with energy penalty, and hill climb on it.

i have always assumed this capability was just a matter of training and that no special architecture is required.

Re:ASM SHADER TOY – It's shader toy but you code in asm on Wednesday June 17, 2026 @09:57PM (#48579633)
It's like jit compression. This should be easy enough to replicate. Is the source of this demo available?
Re:ASM SHADER TOY – It's shader toy but you code in asm on Wednesday June 17, 2026 @09:54PM (#48579617)
thats a nice demo! the text is really cheesy
Re:Ask HN: What Are You Working On? on Saturday June 13, 2026 @06:52AM (#48515875)
I work too much.

We've got: a language that looks like python but lowers to cpp: https://github.com/wegfawefgawefg/dudu; a site that tracks power outtages in canada: https://www.outagehub.ca/ ; a clone of spelunky in cpp with multiplayer and mods: https://github.com/wegfawefgawefg/splonks-cpp; a shadertoy clone but you code in asm: https://github.com/wegfawefgawefg/asm-shader-toy; a fixed point math lib: https://github.com/wegfawefgawefg/gfxp; a sexpr based json replacement with good perf: https://github.com/wegfawefgawefg/gsexp;

Outside of personal stuff, I'd like to shill my consultancy, G&K Software. (IDK if this is even allowed, but its what i'm working on so I think its fair.) We had: an ai asset generation site; a trading tool for a fund in dubai; an endorsed sales data tool;

Those are all out of business now due to global market craziness, tarrifs, and AI advancements... so I guess we've got nothing right now. (Please somebody give us business lol) [email protected]

Re:ASM SHADER TOY – It's shader toy but you code in asm on Thursday June 11, 2026 @10:04PM (#48498982)
My bud was learning assembly but the normal material was pretty boring but I know he wanted to learn shaders so I combined the two into something pretty terrible. It has all the same complexity of normal raymarching, sdf defining, space coordinate warping, except now you dont get variables or functions.

At first it was just a cpp demo with vm that run per pixel. That was slow, but threading it made it plenty fast for whatever peak compute you might need for a shader written in asm. Then added inputs to match shader toy, buffers for feedback and memory, and various inputs, mouse, kb, mic, webcam, etc. After that i looked over the examples and... found them pretty bland. The asm was missing things needed to make it more of an asm and less of a riscv'ish LHA form. So we add features: registers, labels, branches, subr, includes, consts, etc.

Theres a clever two eval pass for consts that is inspired by lisp, where the consts are just the same asm but with a different env. I did it like that because I looked at real asm const DSL's and they look impure to me and violate the intention of the project. Its sort of like comptime if you know what that is and dont know lisp. (What are you retarded?)

I know asm's have lots of funny macro ideas that can make them like pseudo real languages, but I just left that out because at that point why not write a little scheme that targets this "ST-ASM" as an IR, and then a tabbed pseudo python with swizzling that lowers to that. Ill probably do that next for fun.

To share it it had to be on web, and cpu vm per pixel wasnt gonna cut it in js. So the browser version now compiles the assembly to WGSL for WebGPU, with a WebGL2 fallback so Firefox can still run it. (Took longer to get that working than the entire rest of the project, but if it cant run on the top browsers then it can't be shown off...)

Re:Hate Radio on Friday June 13, 2025 @06:38AM (#44267313)
we werent too late tho. hitler lost.
Re:Hate Radio on Friday June 13, 2025 @06:37AM (#44267308)
my personal view is people have a right to be intolerant and hateful and advocate for said intolerance and hate as much as they like.

but property crime and assault are not their right. and those are illegal. so we are already covered there.

Re:Hate Radio on Friday June 13, 2025 @06:35AM (#44267300)
my apologies i was under the assumption that you were defending what I percieve to be infringements on free speech rights.
Re:Hate Radio on Tuesday June 10, 2025 @07:20AM (#44235395)
not false dilema. i only argued against his position by showing that his assertion, even in the extreme worst case such as somebody broadcasting "kill ur bb" would not result in everybody killing ther bb. therefore somebody broadcasting "get rid of browns" does not therefore mean that listeners are brainless zombies who do what they are told will go get rid of browns.

the broadcaster doesnt control my body. i can listen to a broadcast, respect the persons right to assert that we should "get rid of browns", and then simply disagree and think that person is at best ignorant and at worst evil. my capacity to do so is a fundamental requirement for humans to be sentient and participate in a democracy.

if listeners really are the brainless zombies some of the people here advocate for 'protecting' weve got a bigger problem.

if someone broadcasts "kill ur bb" then we should punish those went and killed a bb. the broadcaster has not damaged property or person.

my freedom of speech, whether serious or satire, should not be limited just because there are evil people in the world.

Re:Hate Radio on Sunday June 08, 2025 @02:54PM (#44218853)
enforcing this would require building a panopticon world of brains in jars. no thank you. ill hold individuals responsible for their own actions and keep my freedom of speech thanks
Re:Hate Radio on Sunday June 08, 2025 @02:52PM (#44218845)
no it doesnt. you could blast "kill your own baby" on radio but 99% of people wouldnt do it. and if they did its their own fault.

you either respect the sovereignty of an individual and they are responsible for their actions, or not and if you dont then follow that to its logical conclusion. which would be that all people are not responsible for anything ever, because even the broadcaster was told to by his own life and culture, and so on and so forth until your litigating the first living goo on the planet.

Re:Hate Radio on Sunday June 08, 2025 @02:48PM (#44218823)
most of the world for most of time had slavery. that doesnt mean we should have slavery now. your whole first paragraph is bunk.
Both scarlet/violet sold 27million copies, comparable to all the other mainline games, and legends was the most popular sidegame by far at 15 million copies sold.

I think the series is never going to not be open world going forward. It was a resounding success and most people like it.

i used facebook back when it functioned like that. and it was still retarded then
because my religious redneck family thinks 4chan is a bunch of hackers who hate the elderly because fox news mentioned "ok boomer" meme once.

because ive met many people in real life who think its evil but have never been on it or have even seen it.

every normie hates 4chan, but none of them have never been there.

Re:OpenAI is building a social network on Wednesday April 16, 2025 @10:49AM (#43706299)
the planet is a water closed loop. the vapor condenses as rain somewhere
The general point though is if 99.999% of people doing a thing are fine, but one wacko who likes the thing does a shooting, the evidence for causal connection between the thing and the wackos impetus is beneath the noise floor.

If you are trying to make the argument that The Heros Journey is subtely toxic and evil, well thats just too sophisticated an argument for me. 70% of humanity believes in god. We live beyond truth and measure. Welcome to monkey planet.

I bet you 90% of the haters have never even been to the website.
A healthy skepticism is good.

Have you gone on social media recently? It is like 90% nonfactual weirdness. Even here on hackers news its tons of mutually exclusive unfalsifiable assertions of perspective, not fact.

I dont know about your family, but mine is pretty religious. Listening to their conversation during thanksgiving gives me about a 90% nonfact rate.

I think humans are just are not beings of fact in general.

Do you remember when people thought first person shooters made people into murderers because the colombine guy played doom a lot and made a custom wad to simulate the attack?

If a murderer eats omelettes every day we should ban eggs. Eggs turn people into murderers.

Reminder some kids died jumping off buildings with umbrellas after Marry Poppins. Ban movies.

To add context, every male in my high school went on that site. Pol was just a place crazy people posted. We used to laugh and read eachother dumb copypastas at lunch with gorgonzola cheese rhymes and bad puns.

The average 15yo boy have enough mental hygiene to know everything you read online is false. The website is not a nazi factory.

They have been told by mainstream media that you are supposed to hate it. Thats it. Its totally irrational.
I have leftist friends who grew up looking at memes on 4chan. As adults they remember it fondly.
the serious crime of... deleting egregious posts from a website
Many of the popular internet terms start on 4chan, and then move to reddit and the rest of the internet, and then eventually mainstream news, and 65 year olds mouths. This process takes about 3-5 years.
Re:Google Is Winning on Every AI Front on Sunday April 13, 2025 @02:40AM (#43670600)
"If i never choose, I can never be wrong. Isnt that great?"
My history is not that good, but I have a fondness for the guilded age. Lets look back for wisdom.

Consider a case that was not unique, the growth of iron production in the great lakes area between 1855 and 1865.

In 1855 it was 1000 tons of iron ore. By 1860 it was over 100,000. By 1865 it was a few times greater than that.

Now consider even a single year in there where production is increasing by thousands to tens of thousands of tons. Good business. (The machines used to load and unload those boats and the change in boat designs is awesome by the way. Worth looking up.)

That was early. With much worse technology, and much less capital.

There were crazier deltas in production increase in the 1890's and across the guilded age.

The US natural resources are gigantic. There are 330 million people living there. It has more resources than ever before in history.

Steel and plastic are currently produced in the hundreds of millions of tons per year in the US. That is hardly a middle aged man who cant do a pushup.

With a proper 10 year boom, US production could be exponentially increasing year over year. If it and its people choose it.

A lot of people in the US seem to be envisioning this. It is a really non-abstract vision even for americans of... modest intelligence.

It may be the case that providing the world with banking and facebook, and silicon plans, though possibly much more lucrative than physical production, is just too abstract for the average american to identify with as a positive.

Or it may be the case that physical production is more lucrative than software service export, but that the US government has mismanaged the market constraints in the physical domain and so it just appears to not be the case.

I am not sure which is true. What I do know is that for the average person, the idea of making stuff and trading it is simple to understand, and people like it. Even people who do not make anything identify with this goal. Maybe instead of iron and steel it will be nvidias chips this time around.

I think Americans dont like being the social media export country. Its just not a good future vision you can identify with.

Re:4o Image Generation on Wednesday April 09, 2025 @12:01PM (#43633611)
youre one of todays lucky few. about to have your mind blown. look this one up.
no not that one. the first icm paper:

https://pathak22.github.io/noreward-rl/

and the followup which address the noise impredictability problem.

there are more after that which i believe fail the black pill and miss the point of ml, asicifying the architecture with human priors. But the broader point is to show that rl is not just discovering solutions by chance in random actions. Nature starts with priors, and curiosity is one of the universal policy bootstrapping techniques. (others might be imitation, next state prediction, total nearby replication count)

There is also a paper that deployed ICM on a physical robot and it just played with a ball because it was the only source of novel stimuli, and inadvertantly learned how to operate its arms. There was no other reward in the environment except for curiosity. It is amazing, and slightly creepy. I think the ICM will be rediscovered later in ML tech.

Older »