Mismatched brackets seems like a lower-level job - better done by a linter, not an LLM. Likewise, I wouldn’t rely on an LLM to tell me what properties are on th

Mismatched brackets seems like a lower-level job - better done by a linter, not an LLM. Likewise, I wouldn’t rely on an LLM to tell me what properties are on the response object - that job is better done by e.g. TypeScript. Analogously, “convert this function from Python to lisp” seems like a higher-level job, better done by a programmer human (or at least, it’s too high for the current level of available code LLMs). There is still a valuable level that code LLMs can operate at, in my experience - something like a cross between “a more advanced autocomplete” and “a templating/scaffolding/boilerplate engine”.