I will not be using LLMs in my coding

This report is a continuation of previous blog posts about coding and AI. The author has concluded that they will not use any LLMs or AI agents in their professional work. The decision was reached after considerable introspection and after reading numerous articles and opinion pieces on the matter. The author encourages everyone to read Esther Weidauer's essay titled "Why I Resist Generative AI," which articulates their concerns about AI usage in coding better than they could themselves.

The author acknowledges that they do not entirely agree with all points made in the essay but shares the main sentiment and concerns.

The author highlights several dealbreakers for not using LLMs in their work. Firstly, they do not support the notion that LLMs are just a tool, as they view them as a service where work is outsourced to a third-party company, usually based in the US. This outsourcing raises concerns about control, quality, and liability. The author's background in building production-grade ML implementations and experimenting with locally running LLMs and image diffusion models in a homelab further complicates their stance.

The author discusses the phenomenon of "LLM brain fog," a term used by fellow programmers to describe the disoriented and confused feeling they experience after using LLMs for extended periods. They also mention several studies that confirm this effect. The author argues that LLM output often appears impressive but can be misleading, especially when the generated code contains errors or unnecessary complexity.

This can lead to a loss of literacy and an inability to identify poor-quality code. Additionally, the author notes that LLM-generated code can be difficult to maintain due to the lack of understanding among humans who did not write the code.

The author emphasizes the importance of code maintainability and personal understanding of the code's purpose, goals, and motivations. They feel uncomfortable outsourcing their work to AI providers like OpenAI, Anthropic, or Elon Musk, as these companies may not prioritize the same values or adhere to the same ethical standards.

The author also expresses concern over the environmental impact of new AI datacenters, particularly the deforestation and potential use of fossil fuels in their construction. As an EU citizen, the author is particularly worried about the GDPR implications and the potential violation of data privacy laws when using AI services based in the US.