Unverified60% confidenceFactExact time
GPQA questions are written by domain PhD students and cross-validated, featuring a 'Google-Proof' anti-retrieval design.
2
Sources
60%
Confidence
Long-term
Relevance
7/9/2026
First Seen
Sources
OpenAI's Head of Evals: We've Been Consistently Underestimating What AI Models Can Really Do
bilibiliGelai_AI7/8/2026
Related Claims
Unverified推测性解码由Google Brain于2023年正式提出，核心洞察是GPU在验证一批Token时的计算成本与验证单个Token相近69% similarUnverifiedSimpleQA Verified was introduced by Google DeepMind and Google Research.64% similarUnverifiedGoogle's official documentation recommends using the reverse DNS two-way verification method to verify the authenticity of Googlebot.63% similarUnverifiedAccording to the author's empirical research published in 'The Unshakeable GEO Expert,' after applying the multi-attribute ontology structure to a full algorithmic audit, the case study captured Google AI Overviews' number-one global source card, and the search interface automatically generated an interactive diagnostic table directly based on the chunk data it published, achieving 100% citation monopoly.62% similarUnverifiedMMLU covers multiple-choice questions across 57 subjects, GPQA focuses on doctoral-level science questions and is designed to be Google-Proof, and HumanEval focuses on Python code generation.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/375063API
curl https://kongchang.com/api/v1/knowledge/claims/375063MCP
get_claim(id=375063)