Raw LLM Responses

Because  1. The AI labs are full of programmers, so they’re biased with a programmer-centric worldview 2. Programming is highly measurable; it’s easy to benchmark against, so models are disproportionately better at programming than Freeform tasks like “writing with taste” 3. There is often a “right” or “wrong” answer, so in a world of agents and RL it is easier for the model to test and converge on a correct answer That said, models are a loooong way from realistically replacing programmers at scale. My team at Thunder Compute is hiring more SWEs than ever, which seems to be a trend across SF and more broadly at smaller companies.