Unverified50% confidenceFactExact time
Analysis shows that the frequency of em-dash usage in English text generated by GPT-4 is 3–5 times the average level in human writing.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
8/14/2026
First Seen
Valid until: 11/12/2026
Sources
认知负债：当开发者外包了自己的思考
Related Claims
UnverifiedIn English, 1 token equals approximately 4 characters on average.63% similarUnverifiedWhen asked to write a 5,000-word article, GLM 5.2 actually produced more than 5,000 words, with long-form generation capability far exceeding that of most models.62% similarUnverified研究表明人类阅读字幕时单次注视有效信息量约为7±2个汉字，超过14-16字单行字幕会显著增加认知负荷61% similarUnverifiedDifferent models use different tokenizers (GPT uses tiktoken based on BPE, Claude uses its own tokenizer), and the token count for the same text may differ by 5%–15% across different tokenizers.60% similarUnverifiedThe average information density of spoken language is approximately 40%–60% that of written language.59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/748151API
curl https://kongchang.com/api/v1/knowledge/claims/748151MCP
get_claim(id=748151)