Matt Henderson's banner
Matt Henderson's profile picture

Matt Henderson

@matthen2 • 81,429 subscribers

maths, visualisations, conversational AI. VP Research @polyaivoice prev: @RekaAILabs, @Apple AI/ML, @GoogleAI, PhD @Cambridge_Eng

Shorts

returning to deepdream in the times of LLMs here I am using gemma 4 e4b, starting from a random image and optimizing its pixels to maximise the probability gemma captions them as a fantasy map with mountains, rivers, and castles actual AI art 😅

returning to deepdream in the times of LLMs here I am using gemma 4 e4b, starting from a random image and optimizing its pixels to maximise the probability gemma captions them as a fantasy map with mountains, rivers, and castles actual AI art 😅

33,690 views

what if there was one chicken, with the same mass as all the world's chickens combined? it would be about 1.2km tall. what about one ant from all ant species? one cow? one human?

what if there was one chicken, with the same mass as all the world's chickens combined? it would be about 1.2km tall. what about one ant from all ant species? one cow? one human?

61,885 views

space filling curve

space filling curve

12,044 views

what would a rainbow look like on an alien planet with binary suns? you get two rainbows at different angles - each with a different colour spectrum based on its star's temperature

what would a rainbow look like on an alien planet with binary suns? you get two rainbows at different angles - each with a different colour spectrum based on its star's temperature

56,475 views

uniform random noise + band-pass filter =

uniform random noise + band-pass filter =

34,869 views

I have joined Reka, which today comes out of stealth to announce its Series A. It is an amazing group of people, with a convincing vision for the future of large generative models.

I have joined Reka, which today comes out of stealth to announce its Series A. It is an amazing group of people, with a convincing vision for the future of large generative models.

27,455 views

Videos

LIVE
1.2k

Anya Rossi

sweetdream.ai

SweetDream.aiSponsored•Livecam
Streaming Now

Watch Anya Live

Anya is streaming live right now! Join her private show and enjoy exclusive content.

HD live stream
Exclusive private shows
1.2k viewers online

Free preview available • Premium content

To solve a maze, keep the entrance cold and the exit hot. Once it settles, enter and always walk in the hottest direction.
0:23
matthen2's profile picture

To solve a maze, keep the entrance cold and the exit hot. Once it settles, enter and always walk in the hottest direction.

Matt Henderson

1,607,734 views • 28 days ago

Game of Life, using Minecraft recipes as the replacement rules
0:40
matthen2's profile picture

Game of Life, using Minecraft recipes as the replacement rules

Matt Henderson

235,470 views • 1 month ago

What if our ears were like our eyes, and only had three receptors? Interestingly speech is still more or less intelligible, but not much else too me this shows how much more discriminating our hearing is than our vision
0:48
matthen2's profile picture

What if our ears were like our eyes, and only had three receptors? Interestingly speech is still more or less intelligible, but not much else too me this shows how much more discriminating our hearing is than our vision

Matt Henderson

191,837 views • 27 days ago

Game of Life on Penrose tiles
0:50
matthen2's profile picture

Game of Life on Penrose tiles

Matt Henderson

167,150 views • 1 month ago

what is a multimodal LLM thinking as it watches a video? Gemma 4 12B reads raw image patches, as if they were tokens. It was never trained to predict anything at these 'tokens' - but this video shows what it would predict if you did sample from its next token prediction head
0:53
matthen2's profile picture

what is a multimodal LLM thinking as it watches a video? Gemma 4 12B reads raw image patches, as if they were tokens. It was never trained to predict anything at these 'tokens' - but this video shows what it would predict if you did sample from its next token prediction head

Matt Henderson

163,504 views • 1 month ago

how shaders work- using Excel =LET(x,(COLUMN()-24)/15,y,(ROW()-15)/15,SIN(SQRT(x*x+y*y)*10-t)) Each cell (pixel) computes its colour just from its coordinates- so the GPU can massively parallelise them. The trick is writing the formula to draw the picture you want
0:30
matthen2's profile picture

how shaders work- using Excel =LET(x,(COLUMN()-24)/15,y,(ROW()-15)/15,SIN(SQRT(x*x+y*y)*10-t)) Each cell (pixel) computes its colour just from its coordinates- so the GPU can massively parallelise them. The trick is writing the formula to draw the picture you want

Matt Henderson

112,615 views • 1 month ago

to tell if a maze is solvable, just hang it by its corners! The first maze stays in one piece, so there is no path from the entrance at the top to the exit at the bottom. The second maze splits apart along the solution.
0:35
matthen2's profile picture

to tell if a maze is solvable, just hang it by its corners! The first maze stays in one piece, so there is no path from the entrance at the top to the exit at the bottom. The second maze splits apart along the solution.

Matt Henderson

3,386,663 views • 3 years ago

Deep dreams on modern LLMs are so cool (optimizing an image to maximize P(target caption)) Gemma 12B (left) has no vision encoder — it reads pixels like token embeddings — and stamps recognisable objects around the canvas. E4B (right) has one, and drifts to texture instead.
0:30
matthen2's profile picture

Deep dreams on modern LLMs are so cool (optimizing an image to maximize P(target caption)) Gemma 12B (left) has no vision encoder — it reads pixels like token embeddings — and stamps recognisable objects around the canvas. E4B (right) has one, and drifts to texture instead.

Matt Henderson

95,199 views • 1 month ago

does your audio-input LLM secretly know how to generate speech? yes but it sounds like a scary demon! deepdream but for speech: starting with noise audio, I tried optimizing the probability of a desired transcription, while keeping the model frozen. volume up! 🎧
0:59
matthen2's profile picture

does your audio-input LLM secretly know how to generate speech? yes but it sounds like a scary demon! deepdream but for speech: starting with noise audio, I tried optimizing the probability of a desired transcription, while keeping the model frozen. volume up! 🎧

Matt Henderson

86,422 views • 1 month ago

need to find the greatest common divisor of two numbers? just fold them on top of each other until you can't any more!
0:22
matthen2's profile picture

need to find the greatest common divisor of two numbers? just fold them on top of each other until you can't any more!

Matt Henderson

317,914 views • 6 months ago

one easy way to make a simulation of falling sand piling up- just define what happens for all 16 possible 2x2 grids, and repeatedly apply these rules to the picture
0:50
matthen2's profile picture

one easy way to make a simulation of falling sand piling up- just define what happens for all 16 possible 2x2 grids, and repeatedly apply these rules to the picture

Matt Henderson

1,048,528 views • 2 years ago

searching for Life
0:50
matthen2's profile picture

searching for Life

Matt Henderson

42,015 views • 1 month ago

a version 1 QR code with maximum error correction can store 7 bytes. That’s 2⁵⁶ possible codes. Here are the first 1,043 of them-
0:55
matthen2's profile picture

a version 1 QR code with maximum error correction can store 7 bytes. That’s 2⁵⁶ possible codes. Here are the first 1,043 of them-

Matt Henderson

175,230 views • 6 months ago

an exponential curve looks identical anywhere along the x axis- so long as you stretch the y axis accordingly
0:20
matthen2's profile picture

an exponential curve looks identical anywhere along the x axis- so long as you stretch the y axis accordingly

Matt Henderson

566,786 views • 2 years ago

can you see the spinning dodecahedron? pause the video and it disappears
0:23
matthen2's profile picture

can you see the spinning dodecahedron? pause the video and it disappears

Matt Henderson

303,512 views • 2 years ago

can a neural network learn to walk as a physical object in a physics simulation? here I train walking neural nets with an evolutionary algorithm. The input nodes/feet are activated by sine waves at learned phases & connections between two neurons extend based on their difference
0:49
matthen2's profile picture

can a neural network learn to walk as a physical object in a physics simulation? here I train walking neural nets with an evolutionary algorithm. The input nodes/feet are activated by sine waves at learned phases & connections between two neurons extend based on their difference

Matt Henderson

339,340 views • 3 years ago

space filling curve
0:20
matthen2's profile picture

space filling curve

Matt Henderson

12,044 views • 27 days ago

hello world in python... using a genetic algorithm
0:48
matthen2's profile picture

hello world in python... using a genetic algorithm

Matt Henderson

196,015 views • 2 years ago

Minesweeper, but on the new Einstein tiles
0:31
matthen2's profile picture

Minesweeper, but on the new Einstein tiles

Matt Henderson

241,244 views • 3 years ago

a fun way to draw a picture from the digits of pi, made from circular arcs. draw an arc that goes 3/10 of the way around the circle. Now flip the circle over and continue the curve by drawing another arc 1/10 around, 4/10, 1, 5, 9 etc. following the digits of pi
0:56
matthen2's profile picture

a fun way to draw a picture from the digits of pi, made from circular arcs. draw an arc that goes 3/10 of the way around the circle. Now flip the circle over and continue the curve by drawing another arc 1/10 around, 4/10, 1, 5, 9 etc. following the digits of pi

Matt Henderson

186,931 views • 2 years ago