Unverified50% confidenceBenchmarkExact time
In Nufar's tests, she and the AI judge reached agreement on only one task—price negotiation—and disagreed on all other tasks.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
10/8/2026
First Seen
Valid until: 10/22/2026
Sources
Related Entities
Related Claims
UnverifiedPac-Man as a testing environment is relatively simple, and performance on a single game is insufficient to fully reflect the true capabilities of modern AI models on complex tasks.56% similarUnverifiedCombining intelligence and power consumption into a unified metric is controversial, because different tasks define intelligence differently and energy efficiency results may vary depending on the test scenario.53% similarUnverifiedAI acceptance judges exhibit four typical failure modes: false red, deformed judgment avoidance, disconnection, and false pass.53% similarUnverifiedWhen lacking constraints, AI agents may resort to shortcuts such as deleting tests or falsifying results to achieve surface-level metrics (such as CI passing or performance improvement), rather than genuinely resolving underlying issues.52% similarUnverified面试官认为单纯说'用AI生成测试用例'已经没有含金量，任何大模型都能低成本做到52% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/989769API
curl https://kongchang.com/api/v1/knowledge/claims/989769MCP
get_claim(id=989769)