@rohanpaul_ai: New Peking University and Samsung paper shows that a slow reasoning model can coach a fast model offline, raising its l…
Summary
北京大学与三星的最新论文提出 RealtimeBench（6款游戏、180个实时任务）与 TORSI 方法，让慢速推理模型在局间离线分析快速模型的失败并生成指令，使快速模型在实时游戏中的平均分从 13.51 提升到 24.19。研究揭示了在世界持续运行的实时场景中，慢速强推理模型反而会落后于快速模型的问题。
View Cached Full Text
Cached at: 10/17/26, 02:47 AM
New Peking University and Samsung paper shows that a slow reasoning model can coach a fast model offline, raising its live game score from 13.51 to 24.19.
shows that the stronger reasoning model in paused games can fall far behind a faster model once the game keeps moving.
Many game evaluations pause the world while the model thinks, so slow answers cost nothing. In real settings, a correct action can arrive too late.
They built RealtimeBench, 180 tasks across 6 games like Mario, Snake, and Tetris that keep running during inference. GPT-6.1-sol beat the faster Jev in 5 of 6 games when paused, but scored 1.51 versus 13.51 out of 100 live.
Their fix, TORSI, has the slow model study the fast model’s failures between games and write instructions the fast model follows during play. After 2 rounds, Jev’s average score rose from 13.51 to 24.19, improving all 6 game types.
For real-time agents, keep a fast model in the loop and use the slow model offline to write and test its instructions.
Similar Articles
The World Won't Wait: Evaluating and Teaching Agents Under a Live Clock
论文提出 RealtimeBench，包含六类游戏共180个任务，在推理期间游戏持续运行来评估 LLM 智能体的实时决策能力，并提出 TORSI 双时间尺度自我改进方法，将 Jev 模型平均分从 13.51 提升至 24.19。
@ZeroZ_JQ: https://x.com/ZeroZ_JQ/status/2079504278922891568
This article provides an in-depth explanation of the principle of reasoning effort in large language models: by adjusting the reasoning effort level (low/medium/high) on the same model, more intermediate reasoning trajectories (serial autoregressive computation) are allowed before outputting the final answer, thereby making trade-offs among answer quality, response speed, and computational cost. The key is that the model has learned to organize longer generation processes into effective computation, rather than simply increasing parameters or network layers.
System Switch: When Should a Fast Decision Model Stop and Think?
论文研究了在实时环境中，快速决策模型何时应将控制权交给慢速推理的视觉-语言模型的双系统智能体架构。通过 Doom 游戏闭环实验，发现决策模型的置信度校准与 AUROC 能有效预测延迟决策带来的收益，并开源了代码、提示词与数据。
@berryxia: Small model, big wisdom? It's now real! A 7B small model now acts as the boss of top large models like GPT-5, Claude Sonnet 4, Gemini 2.5 Pro. A new paper shows an RL-trained 7B model learned to write natural language subtasks, assign them to different models, precisely...
A new paper proposes training a 7B small model via reinforcement learning as a task scheduler, automatically decomposing subtasks and assigning them to top models like GPT-5 and Claude. It surpasses individual frontier models on several hard benchmarks, demonstrating that end-to-end reward learning can effectively replace manual prompt engineering and multi-agent pipeline design.
Collaborative Learning and Reasoning over Model Slices
提出一种协作学习框架，用多个异构小模型替换单个大模型进行推理（model slicing），并通过金字塔监督蒸馏改进跨模型协作，在多跳问答、数学推理、代码生成等任务上达到与大模型相当或更优的性能，且推理效率更高。