Bugulet

Some sort of many types of developers

[Bugulet @bugulet1]

Somebody told me it could look like hot chocolate if properly set up, and I tend to agree

[banteg @banteg]

reflexive arcade was a casual games download portal from the 2000s. think bejeweled, diner dash, luxor, mystery case files. they worked with hundreds of devs and distributed thousands of games as shareware versions.

each game was wrapped into a time-limit and registration shell that allowed you to online buy it and keep playing. the service shut down in 2010, leaving the games unplayable behind dead drm.

i stumbled upon a very complete dump recently on rutracker. someone had all the installers saved days before reflexive shut down. it's an impressively complete collection counting 1700 games. the problem with it is it comes with a choose your own adventure 80+ different patches/cracks/keygens and no instructions for which key fits each hole.

the side quest was clear: write the best tools to preserve this collection.

first i made it easy to unpack the games without running the installers. this saves around 13,600 times of clicking "next".

then i discovered most games are merely encrypted behind the reflexive wrapper, so i wrote a static unwrapper, that can strip the now defunct reflexive libraries and pluck the full game out of the encrypted .rwg file.

some games have integrated the reflexive libraries, so there was nothing to unwrap. for these games we needed a keygen.

the wrapper used per-game rsa keys for registration. the public key was embedded in a dll shipped with each game, and the moduli were only 160 bits, which can be cracked on modern hardware in under a second.

so i factored them all and recovered the private exponents for 1,653 titles.

with the recovered keys i built a full toolkit: a keygen that generates valid offline registration codes, a static unwrapper that strips the reflexive shell to extract bare game executables (works for 1,661 out of 1,697 games), and a binary patcher for the remaining integrated wrappers.

the project also includes detailed docs explaining the copy protection mechanism, as well as complete inventory of the collection.

https://t.co/OiVdG8SwqU

[Hedgie @HedgieMarkets]

🦔AI companies are bulk-buying rare books, scanning them through high-speed machines that cut the spines off, and shredding the originals. A service called ISBNdb facilitates orders of up to a million books and keeps buyers anonymous. Pre-2022 books are premium because they're free of AI-generated text. A federal judge ruled the practice is fair use because eliminating the original means only one copy exists at a time. Anthropic hired the former head of Google Books partnerships to obtain "all the books in the world."

My Take
This got to me. A bookseller told 404 Media that rare books with almost no surviving copies are being fed into this pipeline. Books that survived wars, fires, and centuries of handling are being shredded so an AI can learn to write a better marketing email.
ISBNdb's website literally says "'AI company destroys two million books' is not a headline that generates sympathy," and they still built an entire business around making it happen quietly. They offer NDAs as a feature. They coach clients to call it "digital preservation."

I've covered AI companies scraping the internet, torrenting libraries, and stealing music. This is worse because it's irreversible. You can re-upload a website. You can reprint a bestseller. You can't replace the last three copies of an 18th-century botanical text once someone shreds them for training data. And the judge said it's legal. So it's going to accelerate.
"We shred rare books and offer NDAs so nobody finds out" is a legitimate business model in 2026. What a timeline.

Hedgie🤗

[Bugulet @bugulet1]

@mkratsios47 lmao

[Bugulet @bugulet1]

@Nyabinya Yeah it's hacked, the game from the website has been used in a lot of scam sites, original game is called Fountains, a lot of bots make pages with the same stolen assets.

[Bugulet @bugulet1]

@SebAaltonen @IanSamyth might be a great metric, if this took a good couple hundred dollars to make it's barely breaking even but a good benchmark anyways

[Bugulet @bugulet1]

@victormustar so overall how much did this cost?

[Bugulet @bugulet1]

@benjaminvrbk then you need better thinkers in your team

[Bugulet @bugulet1]

@SebAaltonen kinda relevant gdc https://t.co/C7C2wvz81Q

[s&box @s8box]

s&box is now open source — we're pleased to share our .NET 10 game engine source under the MIT license.

Our goal with the play fund, royalty free game exporting and open source are all about giving back to the community so everyone wins.

👉https://t.co/jg8axO2yCP

[bonk @mrbonklet]

corporate executives during game and movie development

[Bugulet @bugulet1]

@ZippyVtuber introspection ain't necessarily baaad
anyways, the rant sounds like nitpicking, using AI to write your frontend is fine (don't know why they were so up about it), it's the backend you have to worry about, and it doesn't look like a complex website to make secure

[Bugulet @bugulet1]

@SebAaltonen @STreffi @toncijukic @ptremblay @DrFuturo_ multiple ways that I tried

1. regular image gen -> 3d model (Hunyuan, TRELLIS)
2. image sequences (video) -> regular compvis techniques to 3D model, or use MoGe/DA v2, etc.

problem was mainly topology but there's some ways to add it to a pipeline (Mesh-RTF/UltraShape)

[Bugulet @bugulet1]

@AlexPalcuie @vtemian do a 5 sigma (figma?) analysis using complicated first order derivatives to figure out when the next outage is coming, easy, i take cash only btw

[Pema99 @pemathedev]

So I was experimenting with a kind of variable rate shading in my toy path tracer today, and doing some questionably valid stuff with unsynchronized memory accesses. Suddenly a sierpinski triangle appears in random garbled mess I've created. What even.... O.O

[Nicolas Dewolf @NicoWolf16]

@cloneofsimo It was my pet peeve when i started research on ML in 2020 and it will remain so for many years to come. Ive even seen papers introduce manifold structures and associated diffgeom notions, to then just proceed with vectors and linear algebra...

[Andrei Kashcha @anvaka]

Map of @Reddit updated: https://t.co/OJi4rAQfXH

Visualizing 116K subreddits from 1.5B comments (Nov 2024-March 2025).

An interactive exploration of community connections across the platform.

I hope you enjoy it♥️

[Pffycat🩷🐾🩵 4.0 Debut Nov 16th!! @pffycat]

Sometimes I just do a random dance when I'm cooking or cleaning or just alone where no one can see me I think it's me defaulting to my idle animations...

[george hotz archive @geohotarchive]

Get Out of Technology https://t.co/btyW0Mx2vR

[anshuman @athletic_coder]

She dumped me last night.

Not because I don't listen.
Not because I'm always on my phone.
Not even because I forgot our anniversary (twice).

But because,

in her exact words:

"You only pay attention to the parts of what I say that you think are important."

I stared at her for a moment and realized...

She just perfectly described the attention mechanism in transformers.

Turns out I wasn't being a bad boyfriend. I was being mathematically optimal.

See, in conversations (and transformers), you don't give equal weight to every word. Some words matter more for understanding context. Attention figures out exactly HOW important each word should be.

Here's the beautiful math:

Attention(Q, K, V) = softmax(QK^T / √d_k)V

Breaking it down:

Q (Query): "What am I looking for?"
K (Key): "What info is available?"
V (Value): "What is that info?"
d_k: Key dimension (for scaling)

Think library analogy:

You have a question (Query). Books have titles (Keys) and content (Values). Attention finds which books are most relevant.

Step-by-step with "The cat sat on the mat":

Step 1: Create Q, K, VEach word → three vectors via learned matrices W_Q, W_K, W_V
For "cat":

Query: "What should I attend to when processing 'cat'?"
Key: "I am 'cat'"
Value: "Here's cat info"

Step 2: Calculate scoresQK^T = how much each word should attend to others

Processing "sat"? High similarity with "cat" (cats sit) and "mat" (where sitting happens).

Step 3: Scale by √d_kPrevents dot products from getting too large, keeps softmax balanced.

Step 4: SoftmaxConverts scores to probabilities:

"cat": 0.4 (subject)
"sat": 0.3 (action)
"mat": 0.2 (location)
"on": 0.1 (preposition)
"the": 0.1 (article)

Step 5: Weight valuesMultiply each word's value by attention weight, sum up. Now "sat" knows it's most related to "cat" and "mat".

Multi-Head Magic:Transformers do this multiple times in parallel:

Head 1: Subject-verb relationships
Head 2: Spatial ("on", "in", "under")
Head 3: Temporal ("before", "after")
Head 4: Semantic similarity

Each head learns different relationship types.

Why This Changed Everything:

Before: RNNs = reading with flashlight (one word at a time, forget the beginning)

After: Attention = floodlights on entire sentence with dimmer switches
This is why ChatGPT can:

Remember 50 messages ago

Know "it" refers to something specific
Understand "bank" = money vs river based on context
The Kicker:Models learn these patterns from data alone. Nobody programmed grammar rules. It figured out language structure just by predicting next words.
Attention is how AI learned to read between the lines.

Just like my therapist helped me understand my focus patterns, maybe understanding transformers helps us see how we decide what matters.

Now if only I could implement multi-head attention in dating... 🤖

Still waiting for "scaled dot-product listening" to be invented.

[GitHub Projects Community @GithubProjects]

Open-source Free Domain For Everyone.