推荐小红书的工作！
Vela: User Behavior as Supervision Sparks LLM Intelligence
Xiaohongshu's Vela uses anonymized user behavior to train LLMs
AISummary
Xiaohongshu's Vela turns anonymized user behavior from real-world scenarios into training signals for LLMs. Its 35B model improves over its base on all 10 benchmarks, averaging +11.5 points on VelaBench and +4.0 points on general benchmarks.
Post on XView on X
Dongxi NLPVerified on X
@dongxi_nlp
1/10 Can anonymized user behavior advance LLM intelligence? Excited to share Vela: turning behavior from real-world scenarios into training signals. Our 35B model improves over its base on all 10 benchmarks: +11.5 pts on VelaBench and +4.0 pts on general benchmarks on average.
Source: Dongxi NLP · x.comPublished · added here