AI Daily – 2025-05-31(Evening)
Stanford University Accidentally Discovers AI Can Generate CUDA Kernels Surpassing Human Experts: A Stanford University research team, while attempting to create synthetic data for kernel generation m
AI Daily – 2025-05-31(Morning)
NVIDIA’s Dilemma and Strategy in “Walking a Tightrope” Between US and Chinese Markets: The Information’s in-depth report reveals NVIDIA’s difficult situation navigating the US and Chinese markets. Jen
AI Daily – 2025-05-30(Evening)
DeepSeek releases new R1-0528 model with significantly improved performance, drawing attention: DeepSeek has launched a new version of its large language model, R1-0528, which has performed exceptiona
AI Daily – 2025-05-30(Morning)
DeepSeek releases R1-0528 model, performance approaches GPT-4o and Gemini 2.5 Pro, tops open-source leaderboards: DeepSeek-R1-0528 excels in multiple benchmark tests including math, programming, and g
AI Daily – 2025-05-29(Morning)
DeepSeek R1’s “Minor Update” Marks a Major Leap with Significantly Improved Programming and Reasoning: DeepSeek released a new version (0528) of its R1 inference model, reportedly with up to 685 billi
AI Daily – 2025-05-28(Evening)
Effectiveness of LLM+RL Training Questioned: Spurious Rewards Can Also Enhance Model Reasoning Ability: Recently, researchers from the University of Washington, Allen Institute for AI, and UC Berkeley
AI Daily – 2025-05-28(Morning)
The Future of RLHF/RLAIF: Can Random/Incorrect Rewards Also Improve Model Performance? : Stella Li’s experiments showed that training the Qwen2.5-Math-7B model with random rewards or incorrect rewards
AI Daily – 2025-05-27(Evening)
Omni-R1: Novel Dual-System Reinforcement Learning Framework Enhances Omnimodal Reasoning Capability: Omni-R1 proposes an innovative dual-system architecture (global reasoning system + detail understan
AI Daily – 2025-05-27(Morning)
Heated discussion on AMD vs. NVIDIA performance in AI inference: SemiAnalysis pointed out testing issues with SGLang on AMD’s ROCm platform, such as deleting failed tests and lowering passing threshol