LLMs are not power tools
Stop comparing large language models to deterministic systems. They are distinctly non-deterministic, more akin to an unruly intern than a reliable compiler.
The persistent mischaracterisation of LLMs as predictable instruments betrays a fundamental misunderstanding. These are not tools to wield with precision; they are probabilistic beasts. Expecting consistent outputs from an inherently stochastic system is a fool's errand. We must abandon the illusion of control and acknowledge their chaotic nature, or face perpetual disappointment. When will we stop projecting our desires for order onto statistical models?
LLMs, despite their current quirks, represent a new class of digital assistant. Their utility, even in probabilistic outputs, far outweighs their unpredictable elements. We would be remiss to dismiss their transformative potential outright.
- ·Misleading metaphors
- ·Deterministic fallacy
- ·Stochastic nature