Skywork R1V2 (38B) now tops 70B+ multimodal systems on major reasoning benchmarks. It's closing the performance gap on OlympiadBench, AIME ’24, LiveCodeBench and MMMU.
The freshest AI/ML research of the weekOur top 7:▪️ Syzygy of Thoughts
▪️ Sleep-time Compute
▪️ Retrieval-Augmented Generation with Conflicting Evidence
▪️ NodeRAG
▪️ DataDecide
▪️ ReTool
▪️ Antidistillation Sampling▪️ Thought Manipulation
▪️ Heimdall: Test-time scaling on the generative verification
▪️ Could Thinking Multilingually Empower LLM Reasoning?
▪️ Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?🧵