XNews

LLM Usage and Prompt Engineering Practices

Several posts discuss the effectiveness and best practices of LLM prompt engineering and usage. A tested prompt ('I am a cognitively challenged PhD student...') shows significant results on GPT-5.1 for literature review, providing clear logic. This prompt is described as capable of explaining the models and empirical methods in papers in simple language. There are suggestions to use a class mcp client to call the mcp server as a best practice in scripts, but questions arise about why not to call LLM directly. The practice of using token consumption or lines of code as KPIs is criticized as simplistic and may lead to counterproductive outcomes.

Key insights

Effective Naive Self-deprecating Prompt on GPT-5.1

A tested prompt starting with 'I am a cognitively challenged PhD student' has been reported to be 'excellent' for reading and explaining academic papers on GPT-5.1. When using this prompt, the model provides clear and structured explanations, particularly excelling in explaining models and empirical methods. This experience comes from direct testing feedback and is recommended for assisting in understanding complex papers.

Using MCP Client/Server in Engineering Integration as a Practice Model

There are discussions suggesting the use of a class mcp client to call the mcp server in scripts as a best practice for inter-system calls or process management. However, some have raised questions about why translation or calling LLM should go through MCP instead of directly calling LLM or scripts, revealing differing preferences and considerations for architectural choices and toolchain design.

Criticism of Using Token Consumption or Lines of Code as KPIs

Some viewpoints argue that using token consumption or lines of code as KPIs reflects a misunderstanding of actual work performance by managers and is a simplistic measurement method. This practice may lead teams to adopt strategies to evade or optimize superficial data around KPIs, ultimately resulting in a lose-lose situation for efficiency and quality. The discussion emphasizes the need to design more appropriate metrics to avoid adverse effects.

Source posts