3 ms·

clearly your original comment was unfair.

• int_19h 3y ago
  Is it, though? The major selling point of coding LLMs is that you can use natural language to describe what you want. If minor changes to wording - the ones that would not make any difference with a human - can result in drastically worse results, that feels problematic for real-world scenarios.
  • visarga 3y ago
    The model is small, so it has weaker semantics.
    • int_19h 3y ago
      I get that. But they are explicitly comparing it to Codex themselves.
• throwaway675309 3y ago
  The criticism stands if you have to continue to rewrite your "prompt" until you can coax out the correct desired output.