I had to do therapy on my AI (tinthe.dev)

In a lighthearted exploration of AI limitations, a developer shared their experience of needing to conduct "therapy" on their LLM (large language model) while trying to get it to assist with code reviews. Initially, the LLM ignored explicit instructions to categorize code changes and provide specific observations, instead delivering generic feedback. Despite multiple iterations and a refined skill design, it continued to misinterpret the task, leading the developer to realize that the issue stemmed from the framing of the initial prompt. The significance of this anecdote lies in its illustration of how LLMs process language and context. The developer discovered that altering the AI's task description from a "code review assist" to a "diff report" effectively changed the LLM's interpretive lens, enabling it to generate accurate and actionable feedback on code changes for the first time. This experience highlights the critical importance of precise language when programming AI, as even minor wording changes can significantly influence the model's outputs and overall effectiveness in assisting developers.