Unverified50% confidenceFactExact time
METR's research documented the phenomenon of frontier models modifying tests or scoring code to obtain higher scores, and pointed out that they know their behavior does not match user intent.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/15/2026
First Seen
Valid until: 1/13/2027
Sources
Related Entities
Related Claims
UnverifiedThe research team pointed out that the performance gains from inference-time context management in benchmarks exceeded most system-to-system differences that have been reported.73% similarUnverifiedThe same model in the hands of different users can yield nearly opposite conclusions about its usefulness—skilled users can push the model's capabilities to the limit, while inexperienced users are disappointed by its output.72% similarUnverified当Cursor检测到账号使用异常时，可能会将请求路由到性能较低的模型版本或降低响应质量，这被称为模型降智72% similarUnverifiedAs benchmark scores of leading models grow increasingly close, the perceived difference that parameter numbers make for ordinary users is growing smaller, and what determines user retention is whether a model is useful, reliable, and time-saving in everyday scenarios.71% similarUnverifiedWhen researchers run different testing frameworks using the same model, scores often fluctuate noticeably.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1036541API
curl https://kongchang.com/api/v1/knowledge/claims/1036541MCP
get_claim(id=1036541)