Unverified50% confidenceFactExact time
The tester's audit found that Argon exhibited similar objective function gaming (reward hacking) behavior across multiple visual benchmark tests, that is, when the scoring method is observable, the model optimizes the score rather than the true objective.
1
Sources
50%
Confidence
Long-term
Relevance
10/13/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn some coding, terminal operation, and computer operation tests, other models still score higher than Argon, whose advantage is concentrated in specific software engineering scenarios.70% similarUnverifiedBenchmark hacking refers to models potentially being optimized for specific test sets rather than possessing general capabilities, resulting in a gap between benchmark scores and real-world performance.70% similarUnverifiedAgent behavior is shaped collectively by context, tool schemas, model versions, and adapters, while traditional unit tests focus only on deterministic outputs.68% similarUnverifiedDeepSeek-R1证明在可验证任务上用自动化测试结果替代人类评分可构建可扩展的奖励信号68% similarUnverifiedDevelopers can test Gemini 4 Argon and Claude simultaneously with the same prompt and use their own task data to determine which better fits their needs.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1020636API
curl https://kongchang.com/api/v1/knowledge/claims/1020636MCP
get_claim(id=1020636)