Co-authored DeepThink with Confidence (DeepConf) paper on reducing LLM reasoning overhead.
How media typically covers Yichao Wang
Referenced in coverage
DeepConf, a test-time method leveraging model-internal confidence signals, improves LLM reasoning efficiency and achieves up to 99.9% accuracy on AIME 2025 while reducing generated tokens by 84.7%.
“Co-authored DeepThink with Confidence (DeepConf) paper on reducing LLM reasoning overhead.”