Unverified50% confidenceBenchmarkExact time
When Jeff is directly asked 'Is this email a phishing email?', accuracy on difficult datasets is close to a coin flip, but accuracy improves significantly when the question is broken down into smaller questions targeting specific features and then combined using code.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/4/2026
First Seen
Valid until: 1/2/2027
Sources
Related Entities
Related Claims
UnverifiedThe accuracy of query results from NL2SQL tools depends on the quality and governance of the underlying data, and vague or ambiguous questions may yield misleading answers.61% similarUnverifiedThe challenge with score normalization is that when a knowledge base contains no truly relevant documents, the highest score after normalization may still be pushed up to the same level as relevant results from other bases, creating an illusion of high scores for irrelevant results.59% similarUnverifiedIntroducing fuzzy semantic matching at the caching layer is essentially trading accuracy for performance, and may return incorrect answers that the user did not request.59% similarUnverifiedPrompt injection exploits a fundamental flaw: the blurred boundary between "data" and "instructions"59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/973501API
curl https://kongchang.com/api/v1/knowledge/claims/973501MCP
get_claim(id=973501)