LLMs are still just low code / no code software (www.marble.onl)

In a thought-provoking article, Andrew Marble reiterates his stance that Large Language Models (LLMs) resemble low-code/no-code software, highlighting significant limitations that may hinder their broader applicability. He argues that, much like low-code tools, LLMs excel in specific use cases but struggle with edge cases and complex tasks that require nuanced understanding and reasoning. Marble suggests that users often blame themselves for LLM shortcomings, such as pattern mismatches or errors in judgment tasks, rather than recognizing these models as limited tools that require careful handling and prompting to function effectively. Marble emphasizes the importance of setting realistic expectations for LLMs, encouraging the AI/ML community to recognize their strengths and weaknesses, similar to how one would approach low-code solutions. This perspective invites developers and organizations to assess the effectiveness of LLMs critically, focusing on productivity metrics, technical debt, and overall quality outcomes. By adopting a clearer understanding of the capabilities and limitations of AI, as with any tool, industry leaders can better navigate its integration into workflows and avoid overestimating its potential.