A recent analysis of five servers and 68 tools revealed that input schemas constitute the largest portion of tool definitions, not descriptions as previously suggested. While descriptions account for about 11.2% of tokens, input schemas make up nearly 30%, and output schemas around 9.5%. The memory server was an exception where the output schema was slightly larger than the input schema, but contrary to prior assumptions, these schemas were author-written and not generated from past runs. This suggests that efforts to reduce definition bloat by trimming descriptions are misdirected, as the primary cost lies elsewhere in the schema definitions. AI
IMPACT This analysis highlights potential inefficiencies in how AI tools are defined, suggesting that optimizing input and output schemas could be more impactful than trimming descriptions for reducing bloat.
RANK_REASON Analysis of technical implementation details regarding token usage in tool definitions. [lever_c_demoted from research: ic=1 ai=0.7]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →