write-ups >

  1. Busy Beavers
  2. RISC vs CISC
  3. SplitCompute and working with Tensors
  4. "Threads cannot be implemented as a library"
  5. Talking to GPUs
  6. The Fortran comeback?
  7. Offline RL Simulation
  8. Interaction Nets
  9. One more ML framework will fix all problems
  10. Quantized LoRA
  11. Contravariance and Profunctors
  12. McKelvey-Schofield chaos theorem
  13. Virtual Water
  14. Forward mode Autodiff and Dual Numbers
  15. Beware of call overheads
  16. Fast array summation with CUDA
  17. Volume of a hypersphere
  18. mmap() and the confusion with LLaMa's sparsity
  19. How did transformers get this good?