BinarySplit @BinarySplit

Joined March 2008

Tweets

58
Followers

12
Following

150
Likes

259

BinarySplit @BinarySplit

10 months ago

@ZIKBAKGURI I saw the first screenshot without context and played all of SOUP ROOMS just to find who made it. It was even better in game. VELDTFUNERAL♡ is something special!

0 0 0 42 0

View Details

@teortaxesTex Likely prior art: Residual Matrix Transformers: Scaling the Size of the Residual Stream, Mak et al., 2025 arxiv.org/abs/2506.22696… Similar implementation (albeit different terminology), similar findings of broad benefit. No MoE/MTP/>1B scaling ablations though.

0 0 0 108 0

View Details

BinarySplit @BinarySplit

a year ago

@ID_AA_Carmack @unixpickle I know, gut reaction: LLMs can't do online learning. But the pieces are coming together for it. TTT makes new weights through online backprop. PEER stores weights in memory. UltraMem scales the memory up.

0 0 0 38 0

View Details

BinarySplit @BinarySplit

a year ago

@ID_AA_Carmack @unixpickle Why try to fix it with backprop though? For realtime, splitting into fast- and slow-thinking models seems inevitable. A HyperNetwork construction like "Learning to (Learn at Test Time)" where the planner edits the actor's weights would obviate the need for online backprop.

1 0 0 72 0

View Details

Patrick Schwab @schwabpa

2 years ago

Understanding human biology across scales - from molecules to cells to entire organisms - remains one of biomedicine's greatest challenges in the fight against disease. Today, we are announcing Phenformer - a multi-scale genetic language model that learns to read and interpret human genomes by connecting DNA, cell and tissue context, molecules and clinical outcomes. Phenformer is a generative model of molecular mechanisms that enables researchers to unravel the mysteries underlying disease, and could thereby accelerate the development of precise future therapeutics.

15 134 696 89K 582

View Details

Wouter van Oortmerssen @wvo

3 years ago

The video for my REAC 2023 talk "Scripting Language? Engine Language? Why Not Both?" is finally online! youtube.com/watch?v=TOnhqo…

4 12 73 10K 20

View Details

Atomontage @atomontage_com

5 years ago

Over two decades of focused work is leading up to the LAUNCH of the first public Atomontage product on the big stage at the largest startup event in Europe: Slush 2021 in Helsinki Dec 1st! Read more about it here: slush.org/product-launch/

3 10 35 0 1

View Details

Trans Lifeline @TransLifeline

6 years ago

Today, we are launching @Peer2PeerLive a robust tool that allows content creators and viewers on Twitch to connect to each other with identity based tags. This project is rooted in the immeasurable value of peer-support and community. Check us out at peer2peer.live

2 45 91 0 2

View Details

authorblues @authorblues

7 years ago

Look at this shit.

5 2 34 0 0

View Details

campfire @forestedheart

8 years ago

important!

62 6K 13K 0 788

View Details

BinarySplit @BinarySplit

8 years ago

@implicit_cast It was one of those workflow optimizing scripts that took hours to conceive and write, but only saved me a few tens of minutes total.

0 0 0 0 0

View Details

BinarySplit @BinarySplit

8 years ago

@implicit_cast I had a very niche use in a userscript: when opening Reddit comment pages in a new tab, if `history.length === 1` it would automatically open the linked content. If I clicked Back, `history.length` would be `2` so it would stay on the comments.

1 0 0 0 0

View Details

Atomontage @atomontage_com

9 years ago

Fantastic technical deep-dive by @EightyLevel on the volumetric quest we're on here at @atomontage_com, with lots of previously unrelased details and imagery: 80.lv/articles/how-v… #voxels #volumetric #inevitable

0 12 36 0 2

View Details

BinarySplit @BinarySplit

9 years ago

@tommulgrew Screen-space normals don't look great if used directly, but if you were only using it to get which octant the normal was in then that's an extra 3 bits of information - the sign bits for each axis. Marching cubes would be better though - easier & more numerically stable.

0 0 0 0 0

View Details

BinarySplit @BinarySplit

9 years ago

@tommulgrew Could you guess the normal based on the surface (generated polygons for the rasterizer / screen-space normal calculation or a marching cubes-style analysis for the SVO ray tracer) and spend the 8 bits refining that initial guess?

1 0 0 0 0

View Details

Azathoth @judges119

9 years ago

every library should have a good science fiction and fantasy section

40 2K 4K 0 6

View Details

Tim Sweeney @TimSweeneyEpic

9 years ago

The story so far: In 1970, Intel invented the microprocessor. This has made a lot of people very angry and been widely regarded as a bad move.

29 1K 4K 0 3

View Details

we're not allowed to say where faun has gone @makoConstruct

9 years ago

Hydranxiety: n. A sense that each problem you try to solve will introduce a great number of subproblems, recursively, forever, until you die

1 6 7 0 0

View Details

BinarySplit @BinarySplit

9 years ago

@mgatland I've been through that too. In theory floats have 25% less precision. In practice, fixed-point invariably wastes 50% of the bits you give it