Q&A to get you started

[elifarley]

How does CEDARScript work?

CEDARScript acts as a "magical librarian" that can analyze and manipulate code based on high-level instructions.

• For code analysis, an LLM can use it to quickly identify and retrieve specific code elements across an entire codebase without having to read every line of code, thus saving you tokens;
• For code manipulation, it allows the LLM to express refactoring intentions using abstract commands, while the CEDARScript runtime handles the low-level details of code modification, such as line numbers, indentation, and syntax.

What are the benefits of using CEDARScript?

1. Improved Performance: Benchmarks show that CEDARScript significantly improves the performance of LLMs in code refactoring tasks, with higher pass rates and fewer errors.
2. Reduced Token Usage: CEDARScript allows for more concise representation of code changes, leading to lower token consumption and cost savings.
3. Increased Efficiency: By offloading low-level tasks to the CEDARScript runtime, LLMs can focus on semantic understanding and higher-level code transformations.
4. Enhanced Code Quality: CEDARScript reduces the likelihood of syntax and indentation errors, resulting in more accurate and well-structured code modifications.
5. Better Scalability: The efficiency and reduced token usage of CEDARScript make it well-suited for handling larger codebases and more complex refactoring tasks.

What languages does CEDARScript support?

Currently, CEDARScript has been tested and proven to work with Python. However, thanks to its usage of Tree-Sitter, it theoretically supports a wide range of languages, including Kotlin, PHP, Rust, Go, C++, C, Java, JavaScript, Lua, FORTRAN, Scala, and C#. Support for Cobol and MATLAB is in development.

How does CEDARScript compare to traditional diff formats?

Compared to traditional diff formats, CEDARScript offers several advantages:

1. Higher Level of Abstraction: CEDARScript allows for expressing code modifications at a more abstract level, focusing on semantic changes rather than character-by-character differences.
2. Improved Robustness: CEDARScript is less prone to errors caused by changes in code formatting or whitespace, making it a more robust solution for code refactoring.

How does CEDARScript impact the performance of different LLMs?

CEDARScript has shown significant performance improvements across different LLMs, including Google's Gemini and Anthropic's Claude. Notably, it enabled Gemini 1.5 Flash, a more cost-effective model, to outperform Claude 3.5 Sonnet, a more advanced and expensive LLM.

How does CEDARScript help LLMs become better "code architects"?

CEDARScript empowers LLMs to act as "code architects" by:

1. Freeing them from Low-Level Details: LLMs no longer need to spend cognitive resources on syntax, indentation, or line counting.
2. Enabling Focus on High-Level Design: They can focus on understanding code functionality, identifying refactoring opportunities, and planning complex transformations.

What are some future directions for CEDARScript?

Future development of CEDARScript will focus on:

1. Embedded Tree-Sitter Query Language: Via CEDARScript, the LLM will be able to interact with the Tree-Sitter parsing library and its query language, enabling more advanced code analysis and manipulation capabilities.
2. Browser Extension Development: Creating a browser extension that allows web-based LLM interfaces to leverage CEDARScript for working with larger codebases.
3. Fine-tuning LLMs for CEDARScript: Training specialized LLM models that natively understand CEDARScript, potentially allowing for even more efficient and accurate code transformations using models that could run on home laptops.
4. LLM-Tool Interface: Exploring the use of CEDARScript as a standardized language for LLMs to interact with various external tools and APIs.