I don’t think LLMs should be trying to replace what essentially should be well tested heuristics.

It’s fine if it’s a stop gap. But, it’s too inconsistent to ever be reliable.