Agent Index
Agent Leaderboards Keep Skewing: High Scores Do Not Prove Capability Gaps
Agent benchmarks face flawed tasks, leaked answers, and sampling error. Judge rankings by confidence intervals, significance, and sample size—not cham
Back to home
SemiAnalysis
2 articles tagged with this topic
OpenAI
OpenAI and Anthropic to Dominate AI Compute; $10T Capex Risks Debt Crisis
Dylan Patel predicts OpenAI and Anthropic will control most global AI compute by 2028; $10T+ capex could trigger a sovereign debt crisis.