write-ups >

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