Idea: generalise iteration by index, rather than by pointer 24ITERATION REVISITED FLUX ▸ Flux (github.com/tcbrindle/flux) is a C++20 library exploring these ideas ▸ Aims: ▸ Much improved safety by default ITERATION OPERATIONS WITH FLUX ▸ flux�::first(seq) returns a cursor, which represents a sequence position ▸ For contiguous sequences, this is just an integer index ▸ flux�::is_last(seq, cur) returns the cursor is past-the-end ▸ flux�::read_at(seq, cur) accesses the element at the cursor position ▸ flux�::inc(seq, cur) increments the cursor 26ITERATION REVISITED FLUX CONCEPTS ▸ A type which provides

of the importers import fmt; fmt.cxx fmt.pcm fmt.oCompilation time #include <flux.hpp> took 319 ms import flux; took 14 msYour imports (can) have imports export module foo; import bar; the gate for a mechanical issue compilation terminated.BMI compatibility (cont’d) add_subdirectory(flux) add_subdirectory(fmt) add_subdirectory(argparse) add_library(foo) target_sources(foo PUBLIC foo fmt::fmt flux::flux argparse::argparse) /lib/llvm-16/bin/clang++ -I/usr/lib/gcc/aarch64-linux-gnu/12/include -I/cxx-modules/cmake-example/fmt/include -I/cxx-modules/cmake-example/flux/include -O3

Training is expensive, serving is getting cheap, and pricing power is slipping. The business model is in flux. And there are new questions about the one-size-fits-all LLM approach, with smaller, cheaper models is expensive, serving is getting cheap, and pricing power is slipping. The business model is in flux. And there are new questions about the one-size-fits-all LLM approach, with smaller, cheaper models