Show HN: Sup AI, a confidence-weighted ensemble (52.15% on Humanity's Last Exam)
Summary
Sup AI presents a 337-model ensemble with real-time logprob scoring, disagreement detection, and lossless context compression, claiming to be the most accurate AI in existence. In Humanity's Last Exam, it achieved 52.15% accuracy, leading by 7.41 points over the next best model, using web search only. The article details the architecture (ensemble search across retrieval methods, per-model prompt adaptation, and extensive transparency features) and argues that aggregated, verified outputs outperform any single model.