Bliki: Paracelsus Maxim
The "Paracelsus Maxim" states that the dosage of any practice, not just its presence, determines whether it is beneficial or harmful Global data is used as a prime programming example: a small amount of immutable global data can be practical, but excessive global state becomes dangerous The article argues against binary thinking about programming habits, advocating instead for contextual and quantitative evaluation The key framework for decision-making is to always ask two questions: "in what co
Analysis
TL;DR
- The "Paracelsus Maxim" states that the dosage of any practice, not just its presence, determines whether it is beneficial or harmful
- Global data is used as a prime programming example: a small amount of immutable global data can be practical, but excessive global state becomes dangerous
- The article argues against binary thinking about programming habits, advocating instead for contextual and quantitative evaluation
- The key framework for decision-making is to always ask two questions: "in what contexts?" and "in what doses?"
Why It Matters
This article offers a nuanced philosophical framework for AI practitioners and software engineers who often face black-and-white debates about best practices (e.g., "global state is bad," "microservices are always better"). Recognizing that the value of any technique depends on both context and scale can prevent dogmatic decisions and lead to more pragmatic, situation-aware engineering choices in AI system design and beyond.
Technical Details
- The Paracelsus Maxim originates from 16th-century Swiss physician Paracelsus: "All things are poison, and nothing is without poison; the dosage alone makes it so a thing is not a poison" (German: "Alle Dinge sind Gift, und nichts ist ohne Gift; allein die Dosis macht, dass ein Ding kein Gift ist.")
- Global data in programming serves as the central case study: limited, immutable global data can efficiently propagate information across a program, while excessive mutable global state introduces coupling, unpredictability, and maintenance challenges
- The article frames the analysis around two dimensions: context (when is a practice appropriate?) and dosage (how much is too much?)
- No benchmarks, datasets, or model architectures are discussed; the piece is a conceptual/philosophical reflection rather than a technical research article
Industry Insight
- AI engineering teams should avoid rigid adherence to architectural dogmas (e.g., "no globals," "always microservices") and instead evaluate trade-offs based on project scale, team size, and system complexity
- When designing AI pipelines or production systems, practitioners should explicitly document the "dosage" decisions they make—justifying why a particular level of coupling, state management, or abstraction is appropriate for their context
- This framework is especially relevant in AI, where techniques like prompt engineering, RAG chunking sizes, and model quantization all follow a similar dose-dependent pattern: more is not always better
Disclaimer: The above content is generated by AI and is for reference only.