Matt Henderson
@matthen.com
Maths animations, dialogue systems research Research @ PolyAI, previously Google, Apple, Reka. PhD Cambridge
1, 0, 2, 3, 4, 5, 6, 7, 8, 9, 10, 12, 11, 13, 16, 15, 14, 17, -1, 18, 24, 20, 19, 21, 23, 22, 30, 32, 25, 36, 28, 31, 29, 27, 26, 37, 40, 35, 42, 48, 33, 34, 41, 64, 60, 43, 45, 56, 44, 47… The most common numbers in the Online Encyclopaedia of Integer Sequences, sorted by frequency
sequence of the smallest numbers that are not listed in the OEIS: 20990, 23543, 23735, 24085, 24159, 24555, 26301, 26673, 26708, 27266, 27815, 27988, 28330, 28353, 28869, 28946, 29577, 29962, 30131, 30146, 30193, 30328, 30466, 30473…
Reaction Diffusion in Photoshop- Blur, Sharpen, Posterize, repeat
The first valid TGA image starts 7.97 million digits into pi (byte 3,310,525 of its binary expansion) It’s 33,207 × 20,901, 24-bit RLE truecolour, so the bits of pi after the header are read as colours and repeat values. Here is its top-left corner-
a parabola is formed when circles radiating from a point meet lines moving at the same speed. this is equivalent to slicing a cone parallel to its slope
To tell if a maze is solvable, just hang it by its corners If it tears into pieces, you’ve found a solution
Seamlessly loops every 10 seconds. But follow any dot and you will be watching for over an hour before it returns to its starting position.
Playing with the P, I, and D weights in a PID controller. P = Proportional: push in proportion to how far the ball is off I = Integral: accumulate historical error to fix slow drift D = Derivative: respond to how fast the error is changing, to damp it
What if gravity fell off as 1/r^1.9 instead of 1/r²? We would have to say bye to the Moon
Drawing all 16,777,216 possible 12 bit RGB colours in one 2D image, each exactly once. Pixels are mapped from the 256³ colour cube, following a three dimensional Hilbert curve, to a two dimensional Hilbert curve filling the square. so nearby colours stay close.
Neat task and training task from Inherent- Mask out the results section of an ML paper, and ask an agent to run new experiments that comes to the same conclusion. Nice test of creativity and understanding- inherentlabs.ai/research/tra...
Training AI Scientists to Replicate Research
We introduce Faraday, an “AI Scientist” agent that outperforms Claude Opus 4.8 and GPT-5.5 on the task of replicating research.
inherentlabs.ai
Late to this realisation, but I have finally given up on Twitter - what a negative and horrible place it has become
