Early Lessons From GPT-4: The Schillace Laws

from:  Early Lessons From GPT-4: The Schillace Laws | Semantic Kernel   John Maeda

https://devblogs.microsoft.com/semantic-kernel/early-lessons-from-gpt-4-the-schillace-laws/

https://learn.microsoft.com/en-us/semantic-kernel/overview/#consider-the-future-of-this-decidedly-semantic-ai

What if you could use natural language to create software? What if you could leverage the power of a large-scale language model that can generate code, data, and text from simple prompts? What if you could balance the trade-offs between leverage and precision, uncertainty and interaction, complexity and simplicity? These are some of the questions that Sam Schillace, a software engineer and entrepreneur, explored when he had early access to GPT-4, the latest version of OpenAI’s generative pre-trained transformer model.

Based on his experience with GPT-4, Microsoft’s Deputy CTO Sam Schillace developed nine principles for using LLMs to create software. We call them the “Schillace Laws”:

1.Don’t write code if the model can do it; the model will get better, but the code won’t.
2.Trade leverage for precision; use interaction to mitigate.
3.Code is for syntax and process; models are for semantics and intent.
4.The system will be as brittle as its most brittle part.
5.Ask Smart to Get Smart.
6.Uncertainty is an exception throw.
7.Text is the universal wire protocol.
8.Hard for you is hard for the model.
9.Beware pareidolia of consciousness; the model can be used against itself.

These principles capture some of the best practices and challenges of using LLMs to create software, especially in domains where natural language is involved or desired. They also reflect some of the strengths and weaknesses of LLMs as a new partner in software development.

 

AI翻译：

倘若你能通过自然语言来创建软件，会是怎样一番光景？倘若你能借助大规模语言模型的强大能力，仅用简单指令就能生成代码、数据与文本，又会带来怎样的变革？倘若你能在效率优势与精准度、不确定性与交互性、复杂性与简洁性之间找到平衡，该有多好？这些正是软件工程师、企业家山姆・希勒斯在提前试用 OpenAI 最新版生成式预训练变换器模型 ——GPT-4 时，深入探究的问题。

 

基于使用 GPT-4 的实践经验，微软首席技术官副手山姆・希勒斯总结出了九条运用大语言模型进行软件开发的原则，我们将其称为希勒斯法则：

 

1. 若模型能代劳，就不必亲自写代码；模型会持续优化，而代码本身不会自我进化。
2. 用效率优势换取精准度，通过交互环节来降低风险。
3. 代码负责处理语法与流程，模型负责解读语义与意图。
4. 系统的脆弱程度，取决于其最薄弱的那个环节。
5. 精准提问，方能获得高质量答案。
6. 不确定性本身就是一种异常信号。
7. 文本是通用的有线交互协议。
8. 对你而言棘手的问题，对模型来说同样不简单。
9. 警惕意识空想性错视；模型的能力也可能被反过来用于对付自身。

 

这些原则既提炼了运用大语言模型进行软件开发的部分最佳实践与核心挑战（尤其适用于需要或偏好自然语言交互的领域），也反映出大语言模型作为软件开发新伙伴所具备的优势与短板。