Unverified50% confidenceFactExact time
A 2024 Stanford University study showed that text generated by GPT-4 differs significantly from human writing on two statistical metrics: perplexity and burstiness.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/7/2026
First Seen
Valid until: 12/6/2026
Sources
Related Entities
Related Claims
VerifiedA 2023 study by Chen et al. at Stanford University confirmed that GPT-4 exhibited significant fluctuations in mathematical reasoning ability over the course of several months.74% similarUnverified2023年多项研究表明，GPT-4在回答医学问题时虚构参考文献的比例可高达30%-70%68% similarUnverifiedMainstream models such as GPT-4, Claude, and Gemini are extremely sensitive to input format and instruction wording, and subtle differences in phrasing can lead to variations in output quality.68% similarUnverifiedA 2023 study found that GPT-4 produced errors in approximately 15–20% of cases when generating code documentation.68% similarUnverifiedLarge language models such as GPT-4 and Claude are already capable of generating articles and news reports that are nearly indistinguishable from human writing.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/871639API
curl https://kongchang.com/api/v1/knowledge/claims/871639MCP
get_claim(id=871639)