Author of the BaiMian large model, offering technology and industry insights.
Google's Gemini 3 and Gemini 3 Pro launch topped major AGI benchmarks such as The Human Last Exam and ARC‑AGI‑2, outperformed GPT‑5.1 in a complex 3‑D gear visualization task, and even generated a functional cloud‑OS prototype, signaling a notable shift toward true artificial general intelligence.
The LightReasoner paper from Hong Kong University shows that small language models can guide large models on critical reasoning steps, achieving up to 90% faster inference and significant accuracy gains across multiple math benchmarks.
UNO‑Bench, an open‑source benchmark from Meituan’s LongCat team, provides the first high‑quality, low‑redundancy unified evaluation framework for omni‑modal large language models, featuring 1,250 manually annotated cross‑modal samples and 2,480 enhanced single‑modal samples covering 44 fine‑grained tasks and five modality combinations.
This article provides a detailed technical analysis of the Olmo‑Thinking project, covering why a new open‑source LLM was built, the challenges of reinforcement learning at scale, data‑mix optimization, architectural bottlenecks such as missing GQA and QK‑Norm, and the post‑training techniques used to improve reasoning and long‑context capabilities.
Kimi's open‑source K2‑Thinking model, a 1‑trillion‑parameter agent with native INT4 quantization and 256k context, achieves SOTA performance on benchmarks like Humanity’s Last Exam, BrowseComp and SEAL‑0, outperforms GPT‑5 and Grok‑4, and demonstrates complex tool‑driven reasoning through real‑world examples.
The author details Kimi Linear's architecture, training challenges, aggressive MoE sparsity, hybrid linear attention design, benchmark gains, and post‑training insights, offering a transparent technical review of this 48B‑parameter MoE LLM built on 5.7 T tokens.
This article analyzes why standard reinforcement learning can degrade Pass@K metrics after fine‑tuning large language models, introduces a risk‑sensitive RL objective that reshapes the advantage estimator, and demonstrates through bandit and mathematical‑reasoning experiments that the RS‑GRPO method consistently boosts diversity and overall Pass@K scores across multiple LLMs.
Meta’s recent paper reveals a sigmoid‑shaped scaling law for LLM reinforcement learning, presents extensive 40‑k GPU‑hour experiments, compares various RL designs such as PPO‑off‑policy‑k and Pipeline‑RL‑k, and distills the findings into a practical “ScaleRL” recipe that improves performance and efficiency.
The article examines the fundamental similarity between SFT and RL loss functions for large language models, explains why RL training is prone to instability, discusses infrastructure and data quality challenges, and reviews practical tricks and reward‑model considerations for more reliable RL fine‑tuning.
The article explains how information entropy, cross‑entropy, and KL‑divergence shape reinforcement learning for large language models, describes the phenomenon of entropy collapse, compares token‑level and policy‑level entropy, and reviews recent methods like Clip‑Cov and KL‑Cov that mitigate this issue.
