Oriented Object Notation – JSON for LLMs at half the token cost

Summary

The article introduces TOON, a compact data format designed to reduce token usage in Large Language Models compared to JSON. TOON achieves this through minimal syntax, indentation-based structure, and tabular arrays. It is not a drop-in replacement for JSON but is designed for LLM contexts where token efficiency matters.

Key Takeaways

1. TOON is a compact, human-readable format designed for passing structured data to Large Language Models with significantly reduced token usage.
2. TOON conveys the same information with fewer tokens.
3. TOON is designed for LLM contexts where human readability and token efficiency matter.

Statement Breakdown

• Claimed Facts: 70% of statements the article presents as facts
• Opinions: 20% of statements classified as editorial or subjective
• Claims: 10% of statements surfaced for additional reader evaluation

Credibility & Bias Reasoning

Credibility assessment: The article provides clear examples and explanations of the TOON format, along with comparisons to JSON. It acknowledges limitations and potential issues, such as tokenizer variations and specific use cases. The claims are generally well-supported by the provided examples and explanations.

Bias assessment: Pro-TOON Efficiency. The article is clearly promoting TOON as a more efficient alternative to JSON for LLMs. While it acknowledges some limitations, the overall tone is positive and focused on the benefits of TOON. The article's primary goal is to advocate for the adoption of TOON.

Note: While the article presents a novel approach, verify token savings with your specific LLM and tokenizer.

Credibility flag: Informative, Consider

Claimed Facts (5)

• This is a factual statement about the design and purpose of TOON.
• This provides a quantifiable comparison of token usage between TOON and JSON.
• This provides a concrete example of JSON data and its token count.
• This provides a concrete example of TOON data and its token count, corresponding to the JSON example.
• This describes the specific formatting rules of TOON.

Opinions (5)

• This is a general observation about the trend of AI development.
• This is a subjective assessment of the benefits of TOON's features.
• This is a subjective description of TOON's syntax.
• This is a subjective assessment of TOON's readability.
• This is a subjective assessment of the potential benefits of using alternative delimiters.

Claims (5)

• While generally true, the degree to which JSON is 'token-expensive' is subjective and depends on the specific use case and tokenizer.
• While this is a rule of TOON, it's presented as a general truth, which is misleading.
• This is a generalization that may not hold true for all tokenizers and LLMs.
• This is a generalization that may not hold true for all datasets.
• While technically true, the practical benefit of this 'self-description' is not clearly demonstrated.

Key Sources

• Author — Article Author