Unverified50% confidenceEventExact time
By comparing two consecutive real request bodies in the same conversation, the team found that the first difference appeared at character 22,966 of the 24,000-character system block (containing the user request's <user_context>).
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/18/2026
First Seen
Valid until: 1/16/2027
Sources
Related Entities
Related Claims
UnverifiedIn the experiment, for two catalogs each containing 40 skills with 1,000 characters per description, the English group was counted as 47,238 characters and the Japanese group was counted as 87,238 characters, exactly 40,000 more.62% similarVerifiedThe vocabulary of a standard autoregressive language model typically contains 32,000 to 128,000 tokens.60% similarUnverifiedAnnotators preferred longer answers, which led the model to learn verbosity, and this is one of the origins of ChatGPT's wordiness.59% similarUnverifiedThe SentencePiece tokenizer used by the Gemma series has a vocabulary of approximately 256,000 entries.57% similarUnverifiedThe study tests the model on two categories of downstream benchmarks: question-answering and semantic text similarity.56% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1057703API
curl https://kongchang.com/api/v1/knowledge/claims/1057703MCP
get_claim(id=1057703)