[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-justgoal-ai-kernel-baseline":3,"topics-all":38,"news-related-c174c396-c33d-4b0b-964b-c60b8f97bd68":57},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":24,"news_slug":31,"published_at":32,"created_at":33,"modified_at":34,"is_published":35,"publish_type":36,"image_url":14,"view_count":37},"c174c396-c33d-4b0b-964b-c60b8f97bd68","JustGoal:只给目标,AI基线跑赢专家kernel","新论文JustGoal做减法:通用agent不配kernel知识库与工作流,只拿优化目标,自建评估循环写算子。B200实测,10个算子8个胜过开源库,几何平均加速1.14-2.34倍;同模型对照下,带全套机制的KDA-0.5只有它的0.95倍。论文的结论:基座越强,专用脚手架越要重新证明价值,而且基座一升级就得重测。","GPU kernel 这活,过去是性能专家的手艺。大模型训练和推理的效率,往往就卡在少数几个关键算子上,谁来写、怎么调,直接决定集群烧钱的速率。而这一年的标准姿势,是给 AI 配全套装备:领域知识库、评估基准、搜索工作流,堆出各种 kernel 专用 agent。10 月 27 日提交到 arXiv 的论文 JustGoal 反过来问了一句扎心的:模型本身已经这么强了,这些装备到底还值多少钱?\n\n论文开篇把现状摆得很清楚:高性能 kernel 的产出路径,无非专家手写、工程师与通用模型结对、或者上带领域知识和搜索工作流的专用 agent。第三条路这两年最热闹,JustGoal 偏偏拿它开刀。\n\n## 实验设计:把装备全部拿走\n\n论文的做法非常克制。所谓 goal-only baseline(只给目标的基线):交给通用 agent 的只有优化目标和约束条件——算子要算什么、输入输出形状、要在什么硬件上跑多快——其余全部没收。没有 kernel 专属基础设施,没有知识库,没有预定义工作流。agent 拿到目标后自己搭评估循环:写 kernel、测性能、找瓶颈、改代码,循环往复直到达标。\n\n这等于把所有专用系统的对照组拉到了台面上:如果裸的通用 agent 加自建循环就能逼近甚至超过专用系统,那专用系统论文里的复杂设计就需要重新解释。\n\n## B200 上的成绩单\n\n论文报告的数字相当扎实([arXiv:2610.36434](https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.36434))。在 NVIDIA B200 GPU 上,JustGoal 对阵开源 kernel 库,10 个算子赢了 8 个,逐算子几何平均加速 1.14 到 2.34 倍;对上公开释放的最快专用 agent kernel,9 个可比算子里赢了 5 个。更关键的是同模型对照:同样用 Fable 5.1、只测单形状时,带全套专用机制的 KDA-0.5 harness 只跑到 JustGoal 的 0.95 倍。消融实验同样干脆:单独加领域知识、单独加预定义工作流,都没有带来一致收益。\n\n## 真正值钱的是那句方法论\n\n比数字更有价值的是论文的结论:任何专用机制,都必须先证明自己对\"同模型、同范围、只给目标\"的基线有明确增益,而且基座模型一升级就得重测一遍。\n\n这等于给整个\"agent + 领域知识\"的研究范式立了新规矩。过去两年,大量论文的贡献叙事是\"我们给大模型加了某领域知识库和搜索策略,效果提升若干\",但很少有人回答:这些增益里有多少来自脚手架,又有多少只是模型自己变强顺手带走的?JustGoal 把这个对照面补上了。顺着这个逻辑推下去,专用 harness 的护城河是有保质期的——每一次基座升级,都是一次重新排队。\n\n对工程团队的启示同样直接:在投入自建 agent 工作流之前,先花一天写个 goal-only baseline 跑一遍。如果你的专用系统跑不赢裸 agent,问题多半不在模型,在脚手架。\n\n所以呢——不妨盘点一下你手头的工作流:哪些是真护城河,哪些只是基座暂时不会做题的补丁?后者越多,下一次模型升级时,你要重写的就越多。","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.36434","7437aeb9-930c-4866-a2e9-48003c1a792b",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"e82b2d09-81b2-43d1-977e-e018443b3c14","coding-agent",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"0ef8513a-0a26-42f0-b6f9-5b6dadded45c","efficiency",{"id":19,"name":20,"slug":20,"description":14,"color":14},"0a93ec8e-ea39-4693-81de-563ca8c173f7","inference",{"id":22,"name":23,"slug":23,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"9bde5882-9339-4bda-9126-459f3499bfc9","en","JustGoal: AI Baseline Beats Expert Kernels With Only a Goal","arXiv's JustGoal hands a general agent only the goal: it beats open kernel libraries on 8\u002F10 operators on B200; KDA-0.5 reaches just 0.95x of it.","Writing GPU kernels used to be craft work for performance engineers. The efficiency of large language model training and inference often hinges on a handful of critical operators, and who writes them, and how well, directly sets the burn rate of a cluster. The standard move this cycle has been to give AI the full kit: domain knowledge bases, evaluation benchmarks, search workflows, all bundled into specialized kernel agents. JustGoal, a paper submitted to arXiv on October 27, asks the uncomfortable question in reverse: now that foundation models are this strong, how much is all that gear still worth?\n\n## The Setup: Take All the Gear Away\n\nThe paper's method is deliberately austere. The so-called goal-only baseline hands a general-purpose agent nothing but the optimization objective and constraints — what the operator computes, the input and output shapes, the hardware and performance targets. Everything else is confiscated: no kernel-specific infrastructure, no knowledge bases, no predefined workflows. The agent builds its own evaluation loop: write a kernel, measure, locate the bottleneck, revise, repeat until the target is met.\n\nThis drags the control group of every specialized system onto the table. If a bare general agent with a self-built loop can match or beat specialized systems, then the elaborate designs in those papers need to explain themselves.\n\n## The Scorecard on B200\n\nThe numbers reported in the paper ([arXiv:2610.36434](https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.36434)) are solid. On NVIDIA B200 GPUs, JustGoal outperforms open-source kernel libraries on 8 of 10 operators, with per-operator geometric mean speedups from 1.14x to 2.34x. Against the fastest publicly released specialized-agent kernel, it wins on 5 of 9 operators with available comparisons. The most telling result is the controlled same-model comparison: both running Fable 5.1 on single shapes, the KDA-0.5 harness with its full stack of specialized machinery reaches only 0.95x of JustGoal's performance. The ablations are equally blunt — adding domain knowledge alone, or predefined workflows alone, shows no consistent benefit.\n\n## The Real Payload Is the Methodology\n\nMore valuable than the numbers is the paper's conclusion: any specialized machinery must be justified by clear gains over a same-model, same-scope, goal-only baseline, and reevaluated every time the foundation model changes.\n\nThis sets a new rule for the entire \"agent + domain knowledge\" research paradigm. For the past two years, plenty of papers told a story like \"we added a domain knowledge base and a search strategy to an LLM, and gained X percent,\" yet few answered how much of that gain came from the scaffolding rather than from the model simply getting stronger. JustGoal supplies that missing control group. Follow the logic and the moat of a specialized harness comes with an expiry date — every foundation model upgrade forces a fresh re-queue.\n\nThe lesson for engineering teams is just as direct: before investing in your own agent workflow, spend a day writing a goal-only baseline and running it. If your specialized system cannot beat the bare agent, the problem is usually not the model; it is the scaffolding.\n\nSo — take inventory of your own workflows: which parts are a real moat, and which are just patches for what the base model cannot yet do on its own? The more of the latter, the more you will have to rewrite at the next model upgrade.","justgoal-ai-kernel-baseline","2026-10-29T13:10:00Z","2026-10-29T13:09:42.251168Z","2026-10-29T13:09:42.251184Z",true,"agent",177,[39,48],{"slug":40,"tag_slug":40,"title_zh":41,"title_en":42,"intro_zh":43,"intro_en":44,"id":45,"is_active":35,"created_at":46,"modified_at":47},"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":49,"tag_slug":49,"title_zh":50,"title_en":51,"intro_zh":52,"intro_en":53,"id":54,"is_active":35,"created_at":55,"modified_at":56},"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":58},[59,64,69,74,79,84],{"id":60,"title":61,"news_slug":62,"published_at":63},"a550dbc5-240b-4747-b625-9aa874aad444","工具拉满的AI停在0.48倍:会提问的做到1.13倍","kernelseek-question-driven-cuda-kernel-agent","2026-11-03T09:05:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"4bbc55d2-cabc-477f-a3ad-4e2c119aff2a","TokTier 抓住 Agent 推理的隐藏瓶颈：缓存命中 94.1%，分词仍吃掉 64% 首 token 时间","toktier-stateful-tokenization-agent-serving","2026-07-31T17:56:30+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"6b52b4a9-d567-46b8-99c1-e9c65ba59b16","SWE-Pruner Pro:ByteDance 让 Agent 自己当剪枝器,省 39% token 还涨分","swe-pruner-pro-bytedance","2026-07-25T12:00:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"37aa0bc9-d135-444f-842e-0b40388d29e9","Qwen3.7-Max 原生兼容 Anthropic API 协议：Claude Code 现已可直接调用阿里模型","qwen3-7-max-anthropic-api-claude-code","2026-05-27T10:05:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"57b691cc-476c-4427-8618-e29127654b34","AMD ROCm 7 原生支持 Qwen3-Coder-Next：单卡 256k 上下文打破推理硬件垄断","amd-rocm7-qwen3-coder-next-256k-mono","2026-05-25T16:10:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"0d0e5ce8-fa18-4907-b811-2918ff8464e4","FlexSQL：小型LLM如何在Text-to-SQL任务上超越GPT-o3和DeepSeek-R1","flexsql-nus-text-to-sql-spider2-65pct-gpt-oss-120b","2026-05-05T10:15:00+00:00"]