陈天桥宣布投10亿美元算力支持发现式智能
Xin Lang Ke Ji·2025-10-30 03:59

Core Insights - Chen Tianqiao announced a $1 billion investment in computational power to support global scientists in innovative AI research [1][5] Group 1: AI's Ultimate Value - Chen Tianqiao believes the ultimate value of AI lies in discovery, emphasizing that discovery-oriented intelligence can actively construct testable world models and propose falsifiable hypotheses [3] - This form of intelligence is characterized by its ability to ask questions rather than just answer them, and to understand patterns rather than merely predict outcomes, thus evolving humanity rather than replacing it [3] Group 2: Five Capabilities for Discovery-Oriented Intelligence - The development of discovery-oriented intelligence requires five key capabilities: 1. Neural dynamic structure for continuous, self-organizing activity [4] 2. Long-term memory for flexible storage and selective forgetting to build knowledge and form hypotheses [4] 3. Causal reasoning mechanisms to infer beyond training distributions [4] 4. World models for internal, unified simulations to predict the future and test ideas psychologically [4] 5. Metacognition and intrinsic motivation systems to drive exploration through awareness of uncertainty and attention control [5] Group 3: Support for Young Scientists - To assist global scientists in advancing discovery-oriented intelligence research, several initiatives targeting young scientists were announced: 1. Benchmark testing that evaluates across neural dynamics, memory, causality, world models, and metacognition with discoverability as a core metric [5] 2. Structural computational power investment of $1 billion to prioritize support for structural experiments [5] 3. A PI incubator providing independent pathways for PhD students and postdocs to establish their own labs and lead teams without traditional timelines, alongside the establishment of global R&D centers [6]