-
Personal Agents and Memory 
Liao Jiayi #AI#LLM#Agent A survey of Agent Memory techniques and industry practice—from external memory and in-model state to parameter updates—and how memory can make agents intelligent and personalized.
-
DeepSeek-V4.1 Data Recipe and Key Features Liao Jiayi #AI#LLM#Data In the Post-Training section of its Tech Report, DeepSeek-V4.1 emphasizes that its gains come almost entirely from training data. With internet data already exhausted, how can we find better data?
-
Agent Model: Nemotron - Data Recipe Liao Jiayi #AI#大模型#Data Great open-source work from NVIDIA: Training and Data are both fully open source. Well worth a look.
-
Continual Learning: The Next Direction for Intelligence? Liao Jiayi #AI Continual learning lets models absorb feedback during use and keep improving, adding another path to intelligence beyond pretraining and post-training/RL.
-
Tracing Large Models in Recommendation Systems Liao Jiayi #AI#Recommendation systems#LLM New wine bottles for familiar ideas: how can LLM techniques systematically reshape recommendation systems?