The article challenges the common criticism that LLMs' nondeterminism is their fundamental flaw, arguing that the real issues lie elsewhere. It explains that while compilers are deterministic functions producing consistent outputs from given inputs, LLMs intentionally incorporate randomness for creativity. The author contends that blaming nondeterminism oversimplifies the challenges of using LLMs in production systems.
Background
Large Language Models (LLMs) are often compared to compilers in discussions about their role in software development, with debates about whether their outputs should be trusted like compiler outputs. The deterministic nature of compilers versus the often non-deterministic outputs of LLMs is a key point of contention in these discussions.
- Source
- Lobsters
- Published
- May 5, 2026 at 04:38 AM
- Score
- 7.0 / 10