I Don't Want to Code with LLM's
8 hours ago
- #LLM
- #Software Development
- #Productivity
- The author expresses skepticism about the productivity gains claimed from using LLM (Large Language Model) coding tools, citing a study that showed a 19% productivity loss among developers using these tools.
- While acknowledging that LLMs can handle small, well-defined tasks like file conversions or simple data mapping, the author considers these gains minor and comparable to existing IDE features.
- The author argues that LLMs are not suitable for complex coding tasks that require deep understanding of the system, as they often break down when faced with intricate problems.
- Reviewing code generated by LLMs is seen as less effective than writing code oneself, as the process of writing helps in structuring thoughts and building a deeper understanding of the subject.
- The author warns against the atrophy of coding skills due to over-reliance on LLMs, drawing parallels to how reliance on calculators has diminished arithmetic skills.
- There is a critique of the shift in developer culture from valuing deep technical knowledge and mastery to prioritizing the ability to produce code without deep understanding, facilitated by LLMs.
- The author concludes by stating a deliberate choice to avoid using LLMs in their software development career, valuing the process of coding and understanding over potential short-term productivity gains.