LLM-assisted coding is not deterministic. It does it matter?
AI summary of the linked articleThe article argues that determinism and predictability are different properties, and that for LLM-assisted coding the useful question is which workflow yields more predictable outcomes, not whether the output is deterministic. It explains that a system can be deterministic yet unpredictable, as with weather, where physics is deterministic but measurement limits make forecasts hard, and that software behavior is similarly hard to foresee given the complexity of hardware, libraries and configurations. It notes that end users mainly care whether software behaves correctly most of the time and whether failures are visible and fixable. It cites DO-178C, the aeronautics certification standard, as an objective-based framework that could apply to both human coders and LLMs. |
Comments No comments yetWhat caught your eye? Start the discussion. |
More from blog.vrypan.net on Kiwi News |
