What if useful AI is a fantasy? (lzon.ca)

A programmer's exploration of AI tools like Claude and Claude Pro has led to mixed results, raising questions about the actual utility of AI in development work. Initially, the AI-assisted coding experience was exhilarating, allowing for rapid project completions. However, as the programmer grappled with debugging and feature enhancements, it became clear that without a strong mental model of the generated code, the initial excitement faded into frustration. This journey highlights a growing trend among developers to lean toward AI-assisted ideation and guided programming, rather than relying on AI for complete code generation. The significance of this narrative lies in its reflection of a broader sentiment within the AI/ML community regarding the limitations and expectations of large language models (LLMs). As the programmer contemplates the future viability of LLMs, drawing parallels with past technologies like the Metaverse, it raises crucial questions about their long-term value. Will the industry ultimately find a "killer app" for LLMs, or will they be viewed as an overhyped technology that falls short of delivering substantial benefits to developers? This ongoing contemplation emphasizes the need for continued exploration and adaptation as AI evolves.