WHY I'M STILL BEARISH ON LLMS AFTER NAVIER-STOKES

Summary from the TLDR Dev newsletter

Frontier models can produce great results on tightly specified tasks, but most knowledge work lacks cheap, rigorous verification. The argument is that human oversight and specification costs will keep LLMs closer to fast, capable interns than autonomous replacements, while cheaper open models may win many practical workloads.