Unverified50% confidenceFactExact time
Despite training data coming from machine learning and algorithmic programming, AREX-2 can transfer to deep research tasks, supporting the assumption that capabilities are domain-agnostic
1
Sources
50%
Confidence
Long-term
Relevance
10/1/2026
First Seen
Sources
Related Entities
Related Claims
Unverified大模型研发工程师需掌握深度学习算法、PyTorch与TensorFlow、大模型微调、HuggingFace生态、强化学习、MoE训练等，但不需要深入最底层数学推导70% similarUnverifiedBreakthrough work often appears at the intersection of two fields: FlashAttention sits between deep learning algorithms and computer architecture, while Muon sits between optimization theory and deep network training practice.69% similarUnverifiedThe chips DeepSeek is developing in-house are aimed at inference rather than training.69% similarUnverifiedOpponents argue that existing deep learning systems are essentially statistical pattern-matching tools and are far from architectures capable of forming autonomous goals69% similarUnverified现代深度学习方法采用序列到序列（Seq2Seq）方式直接从图像特征预测字符序列以绕过字符切分瓶颈，但对训练数据需求量更大68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/963926API
curl https://kongchang.com/api/v1/knowledge/claims/963926MCP
get_claim(id=963926)