LLMs Are Not a Higher Level of Abstraction
8 hours ago
- #Determinism
- #LLM
- #Programming Abstraction
- The claim that LLMs are the next step in programming abstraction, similar to moving from binary to assembly to C to Python, is incorrect.
- Previous layers in the tech stack involve deterministic functions where a specific input always yields the same output, while LLMs produce probabilities of outputs, not guaranteed results.
- LLM outputs include not only the desired artifact but also unintended elements, potentially introducing security risks or unwanted features.
- The author emphasizes the need for self-aware programmers rather than those who uncritically accept AI-generated artifacts.