LLMs are not a higher level of abstraction (www.lelanthran.com)

A critical perspective on the characterization of Large Language Models (LLMs) as a higher level of abstraction in programming has emerged, disputing the notion that LLMs represent a logical progression akin to the evolution from binary to assembly, and then to higher-level languages like C and Python. The argument presented emphasizes that traditional programming languages operate on deterministic functions where specific inputs yield predictable outputs. In contrast, LLMs produce probabilistic outputs, where a single input generates multiple possible results rather than a single artifact, leading to unpredictable and potentially harmful outcomes. This viewpoint is significant for the AI/ML community as it highlights the limitations of LLMs, especially in critical applications where precision is paramount. The authors caution against accepting LLMs' outputs as solely reliable, as they may incorporate unwarranted artifacts that could pose safety risks. As the use of LLMs proliferates, this critique encourages developers and practitioners to cultivate a deeper understanding and vigilance regarding the outputs generated by these models, urging them to be critically aware rather than passive conduits of AI-generated content.