[drra]

The notion of this article is that LLMs are just for answering question as if these were failed oracles. LLMs are excellent at instruction following to the point where it'll execute instruction by creating bullshit to fill the gap. It's unlikely to get a perfect C code out of Starbucks barista but that doesn't mean these people are in any way flawed at their job. LLMs given enough information about a task can perfectly execute it and enough is far less than any of systems created before.