Benchmarks Versus Workflow
Alex Mashrabov tells 20VC that prompting and workflow are too complex for benchmarks to represent, then returns to model selection. The written definition is cooler: selection picks a candidate by a performance criterion. Occam's razor asks for the smallest set of elements, but Mashrabov's criterion includes operators, prompts, and production flow. His skepticism is not anti-measurement; it is a demand for measures that survive work.






