AI Efficiency Hub

OpenAI suspects DeepSeek of using its models for training. Model distillation is at the heart of the debate, sparking market reactions.

DeepSeek has made progress in open - source large models. The open - source vs closed - source battle continues, and commercialization will accelerate in 2025.

DeepSeek-R1 shocks Silicon Valley. It's a 'byproduct' of Magic Square Quant, succeeding through innovation, open source, etc.

DeepSeek-R1, an open-source deep reasoning model, has unique features and its low cost may impact AI adoption.

DeepSeek's founder Liang Wenfeng shares views on price war, innovation, talent, culture, AGI ambition, competition and collaboration in AI.

DeepSeek's Liang Wenfeng discusses various aspects like price war, innovation, talent, and the future of AI model research.

UI-TARS is a free AI by ByteDance and Tsinghua, with great capabilities and innovative training.

2025 年 01 月 26 日 08 点 56 分 11 秒，星期日

AI's essence is a human attempt to combat entropy. It's a blend of science and humanism.

A challenging large model test set with over 3000 graduate-level questions. Models struggle to score above 10%.