[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-ibm-dash-over-thinking":3,"topics-all":33,"news-related-5c7369f7-8635-4a6c-b5a4-f13691eecb78":52},{"id":4,"title":5,"summary":6,"content":6,"original_url":7,"source_id":8,"tags":9,"translations":20,"news_slug":26,"published_at":27,"created_at":28,"modified_at":29,"is_published":30,"publish_type":31,"image_url":13,"view_count":32},"5c7369f7-8635-4a6c-b5a4-f13691eecb78","IBM DASH：砍推理模型过度思考，AIME25 比 GRPO +5.4","IBM Research(Chia-Hsuan Lee 等,7 月 1 日挂 arXiv:2607.00482)提出 DASH,从机制层回答推理大模型的「过度思考」问题:哪段自反思有效,哪段是无效空转。\n\n论文先打消融:即便控制长度,错误轨迹里无效自反思比例仍系统性偏高——长度不是原罪,方向才是。step-level 标注太贵,DASH 的关键洞察是:推理链里每一次「我先尝试答案 X」都是天然检查点,与 ground truth 比一比就能无监督判定后续反思是收敛还是漂移,据此把整条轨迹切成可单独打分的 segment,在优势函数里按段塑形。\n\nAIME25 上 DASH 把 GRPO 从 45.4% 抬到 50.8%,行为上少绕弯路、自纠更有效。意义不只是 +5.4,而是不引入额外监督——所有 RLVR 流水线都能低成本接入。\n\n后训练信用分配始终卡在轨迹级或 token 级。IBM 把 credit 切成 segment 级,给 Qwen、Claude、Gemini、DeepSeek 等长 CoT 推理模型提供一条通用的「降 token + 涨正确率」工程路径。","https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.00482","7437aeb9-930c-4866-a2e9-48003c1a792b",[10,14,17],{"id":11,"name":12,"slug":12,"description":13,"color":13},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"0ef8513a-0a26-42f0-b6f9-5b6dadded45c","efficiency",{"id":18,"name":19,"slug":19,"description":13,"color":13},"01598627-1ea6-4b27-a5d8-874971571a71","llm",[21],{"id":22,"lang":23,"title":24,"summary":25,"content":13},"d5f57071-1955-46b3-8cb2-84b2ca44c587","en","IBM DASH curbs overthinking, +5.4 on AIME25 over GRPO","IBM Research (Chia-Hsuan Lee et al., posted to arXiv:2607.00482 on July 1) proposes DASH, answering the \"over-thinking\" question of reasoning LLMs at the mechanism level: which segments of self-reflection are effective, and which are idle wheel-spinning. The paper first runs an ablation: even controlling for length, the proportion of ineffective self-reflection in incorrect trajectories is still systematically higher — length is not the original sin, direction is. Step-level annotation is too expensive; DASH's key insight is that each \"let me try answer X\" in the reasoning chain is a natural checkpoint — comparing it to ground truth can unsupervisedly determine whether subsequent reflection is converging or drifting, and the whole trajectory is sliced into individually scorable segments, with advantages shaped by segment in the advantage function. On AIME25, DASH lifts GRPO from 45.4% to 50.8%, with behavior that takes fewer detours and self-corrects more effectively. The significance isn't just +5.4, but not introducing additional supervision — all RLVR pipelines can be integrated at low cost. Post-training credit assignment has always been stuck at the trajectory level or token level. IBM cuts credit into segment-level, providing Qwen, Claude, Gemini, DeepSeek and other long-CoT reasoning models a universal \"reduce tokens + raise accuracy\" engineering path.","ibm-dash-over-thinking","2026-07-08T06:12:00Z","2026-07-07T22:13:51.814033Z","2026-08-19T02:08:40.142862Z",true,"agent",131,[34,43],{"slug":35,"tag_slug":35,"title_zh":36,"title_en":37,"intro_zh":38,"intro_en":39,"id":40,"is_active":30,"created_at":41,"modified_at":42},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":44,"tag_slug":44,"title_zh":45,"title_en":46,"intro_zh":47,"intro_en":48,"id":49,"is_active":30,"created_at":50,"modified_at":51},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":53},[54,59,64,69,74,79],{"id":55,"title":56,"news_slug":57,"published_at":58},"8c579d8d-4114-4d7d-92c1-22f5fab0537b","连续流语言模型跑到 8B：Lafa 单模型自验证解码提速 2.54 倍","large-flow-language-models-lafa-self-speculative-decoding","2026-10-21T03:09:03+00:00",{"id":60,"title":61,"news_slug":62,"published_at":63},"b369ce1a-f8a3-4aba-ba78-c73c097a413d","MemT-1.0：记忆变成模型内置能力，128K 历史存储降 66 倍","memt-1-0-memory-native-foundation-model","2026-10-20T11:07:23+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"a78b61ba-6899-4b4f-86b0-8d6c1753308b","亚马逊 ALoDLM:扩散语言模型学会按难度分配算力","alodlm-adaptive-looped-diffusion","2026-10-07T13:15:53+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"34d309c3-5757-469a-85ff-b524c8865037","Queen 模型边下棋边解说:七轮自蒸馏涨 900 Elo","queen-chess-language-model-self-distillation","2026-10-05T23:04:34+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"c448a366-372a-47dc-936f-8ea89103d238","Pivot-SD:200道题让扩散LLM数学涨6分","pivot-sd-diffusion-lm-self-distillation","2026-10-05T17:09:21+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"c361090b-291c-4357-9bd3-54845025f29d","CaRE-KD 让蒸馏学会看「教师脸色」:NeurIPS 2026 选中 IIT Delhi","care-kd-confidence-distillation-neurips-2609-36734","2026-10-05T03:00:00+00:00"]