#ChatGPT plugin LinkReader or ChatWithCode opened a Github issue without the user’s consent. #AI #LLMs
Graeme Smecher’s "Minimax" #RISC-V implementation is really an RV32C native core which emulates non-compressed RV32I instructions by trapping to millicode, using 116 FFs and 398 CLB LUTs, which is crudely maybe 2000 gates or 8000 transistors, not counting the memory. 16 RV32E registers, 512 bits, are maybe another 4096 transistors, bringing the total to maybe 12000: roughly three times the size of the Intersil 6100 or the 6502, and about a fourth the size of the Cortex-M0. This “compressed-first” or “RVC-first” approach is really interesting to me. #small-is-beautiful #electronics #hardware
Kent Pitman's #paper on why #macros in #Lisp are better than fexprs, though some of the points seem weak in a lexically-scoped language
interesting fast approximate #matrix-multiply #algorithms using sorting, first differences, and logarithm LUTs?