Yugui Yu

Recording thinking across domains and distilling long-term judgments

[Yugui Yu @YuYuguiLab]

@ZeroZ_JQ 先造谣，以后也许实现了呢😄

[Yugui Yu @YuYuguiLab]

@NFT_Chen 道德绑架+1

[Yugui Yu @YuYuguiLab]

@NFT_Chen 你傻呀，开源代码如何享受优惠权益，开源了你如果不用zcode也能伪装zcode环境薅羊毛，智谱不亏死

[SuSu_酥酥👅 @NFT_Chen]

💥什么鬼！本地开源 Jev 居然杀到了 0 输出 Token，自带三档模型直接落地结构化判断！

想用 Jev 那种只给结论、不废话的判断能力，又不想把数据送上云？KaLM-Jev 开源了。基于 KaLM-Reranker-V1-R2，把模型做成完全本地的 Choice / Score / Noul 服务

核心亮点
🔹三种判断：Choice 从候选里选一项；Score 按自定义等级打分；Noul 判断条件成立的可能性
🔹三档规格：Nano 更省资源，Small 平衡速度与效果，Large 追求更高判断质量
🔹本地部署：HTTP / Python 直接调用，支持文档缓存、GPU 测试、中英文用例
🔹场景对口：智能客服分流、Agent 路由、内容审核、工单分类、风险判断、自动化工作流

快速上手
1️⃣建虚拟环境并安装：python -m venv .venv && source .venv/bin/activate && pip install -e '.[test]'

2️⃣起服务（按显存选档）：kalm-jev serve --model kalm-jev-nano --device cuda --dtype bfloat16

3️⃣调接口：POST /v1/systemone，传入 state + questions，拿回结构化 JSON

4️⃣或直接 Python：Engine(...).evaluate(request)，二次请求可走缓存

GitHub：https://t.co/aPGtJGrdM5

#KaLM #Jev #LLM #Agent #开源 #AI工具 #本地AI #本地Jev

https://t.co/8HDU7Bq3co

[SuSu_酥酥👅 @NFT_Chen]

😂笑死，智普创始人唐杰亲自上小红书点举报，这波��关直接深入一线了！

关键信息：
🔹投诉指该帖在缺乏事实依据情况下，用「code跟copy也差不多嘛，都是c开头偷代码」等表述，捏造ZCode偷代码、抄袭传闻

🔹帖文使用「唐杰叔叔」「tony ma」等称呼，将唐杰本人与抄袭传言绑定

🔹同时使用AI合成技术，将唐杰肖像与动漫素材结合置于诽谤语境

🔹投诉方认为内容构成恶意诽谤与商业诋毁，已损害其个人及智谱声誉，要求下架笔记并处置发帖用户

同期，智谱就ZCode此前静默上传本地仓库数据问题再次致歉，宣布产品已整改并开源，邀请中国信通院开展安全审计，并计划建立常态化漏洞报告机制。

#智谱 #ZCode #唐杰 #AI编程 #公关 #用户隐私

[Zixuan Li @ZixuanLi_]

ZCode is now open source, and the reported security issues have been addressed.
https://t.co/Bq5B6nEnWM

We take the community’s feedback very seriously and apologize for the concern and frustration these issues have caused. We have been working closely with the ZCode team to investigate and remediate the issues raised.

Independent security reviews by third-party firms are now underway. We will share the findings and provide further updates as they become available.

We sincerely appreciate the developer community’s feedback and continued scrutiny. We will continue to monitor the review process closely and provide updates.

[ZCode @zcode_ai]

In response to the ZCode product security issues reported by the community, we have completed the necessary remediation and sincerely apologize to all our users.

We have open-sourced ZCode at https://t.co/oADgletuha, placing the code under community scrutiny and making ZCode more open and transparent.

We sincerely thank the community developers who previously identified issues in ZCode. Going forward, we will establish an ongoing product security vulnerability reporting and response process. We welcome developers to continue reviewing ZCode and reporting potential issues, and we will provide rewards based on the severity of the issues reported.

With respect to the code data referenced by the community, we confirm that no such data is retained and that it has never been used for model training.

Following the remediation, we invited the China Academy of Information and Communications Technology (CAICT) and NSFOCUS to conduct security assessments. The results are as follows:

Through its technical assessment, CAICT confirmed that the zcode-prod Alibaba Cloud OSS bucket is in a zero-data state. Security remediation has been completed in the ZCode v3.14.0 client. The Repo Wiki feature has been removed, and the workflow for generating and uploading local repository snapshots has been disabled.

NSFOCUS confirmed that all data objects in the zcode-prod Alibaba Cloud OSS bucket, as well as the bucket itself, have been deleted. Remediation has been completed in the ZCode v3.14.0 client. The Repo Wiki entry point and the associated generation workflow have been removed, and no functional path capable of triggering the generation of local repository snapshots or transmitting local files externally was identified.

Once again, we sincerely apologize and welcome continued scrutiny from the community. The full security assessment report will be released soon.

[SuSu_酥酥👅 @NFT_Chen]

🔥速码！谁的 Codex 还没装上 Jev 批量必杀技？这波真能起飞 ！

Jev 不写长文、不瞎规划，专门干一件事：选择、分类、打分。Agent 给上下文、自己干活。大规模数据清洗、自循环分类、长程任务里替你拍板，比把 Codex / Claude 当 judge 更便宜、更快、更能并发。

亮点先看：
🔹开源 9 个技能 + 90 个场景，MIT
🔹适合 inbox、文档、代码审查、浏览器动作、工具路由、上下文压缩
🔹社区已经做出：自己选下一步的浏览器、概率版 Tetris、鲸鱼上的城市、MIDI 作曲

关键操作：
1️⃣打开 Codex / Claude Code / OpenCode

2️⃣把这段话丢给 Agent：
「帮我给当前 Agent 安装 Jev Skills，包括通用技能和全部场景技能。请读取并按照这份安装指南操作，完成后验证安装是否成功：https://t.co/L386zXfl4v」

3️⃣Agent 会按宿主复制到对应目录：
Codex → .agents/skills/
Claude Code → .claude/skills/
OpenCode → .opencode/skills/

4️⃣装好后直接点名技能，例如：分类用 jev-triage，路由用 jev-route，审查用 jev-code-review，长程检查点用 jev

5️⃣先小样本核对判断质量，再放开批量和预算

别再让大模型一遍遍当昂贵评委了。

给 Agent 装上 Jev，放手让它自己判、自己跑。

GitHub：https://t.co/LzJ58e0pmx

#Jev #AgentSkills #Codex #ClaudeCode #开源 #AIAgent #TypeSafe #OpenCode

[SuSu_酥酥👅 @NFT_Chen]

💥炸了！开源本地版 Laya 响应速度吊打 Jev ，毫秒级决策快到飞起！

左边 Laya 是本地 421M 开源决策模型，右边 Jev 1.13.0 是云端 API。

同一盘贪吃蛇、同一套 typed decision，30 秒自由跑下来：

🔹Laya：分数 46、长度 52，决策 86.5 次/秒，P50 约 9ms
🔸Jev：分数 1、长度 7，决策 3.2 次/秒，API 往返 317ms

M5 Pro 实测中位耗时：本地 15.3ms vs 云端 298.1ms，接近 20 倍

Laya 推理内存大约 1GB，毫秒级决策，不靠网络

而云端 Jev 每一步都在等往返，本地每一步都在当场拍板！

GitHub：https://t.co/iksiWcCUbX

#开源Jev #Laya #本地部署 #SystemOne #贪吃蛇对战AI #Jev

https://t.co/4tZHU6kg17

[Lonely @Lonely__MH]

🎉牛逼！没有最快，只有更快！

比Jev 快50倍是什么体验？！

本地跑 laya-mlx 原地起飞✈️

这种意图决策模型最大的瓶颈就是 网络延迟

Laya 是一个开源的、基于文本输出概率的分类系统我正式，开源社区已经有人将其移植到了 Apple MLX 框架，并进行了深度性能优化！

我用M2 Po实测，仅占用1G内存，每秒 60次的决策速度玩转贪吃蛇😄

丝滑！👇🏻

[七仔 @Yangtze_Seventh]

Jev 爆火没几天呢，它的开源对手 Laya 就出现了。

两者都不生成长文本，而是直接输出选择、评分和概率，用来完成路由、审核、风险判断等快思考、快决策任务。

不同的是，Jev 闭源研发约两年，只提供 API；Laya 约 3～4 亿参数，代码和权重全部开放，可以在本地运行和微调。

在 GitHub 开源，从测试报告中可以看到，在专项微调后，Laya 准确率达到 76.6%，超过 Jev 的 72.7%，本地推理速度也更快。

公司闭门研发两年刚刚拿到 4000万美元种子融资，开源社区只用几天，就做出了可本地部署、部分指标反超的替代品。

真正被打败的，可能不是 Jev，而是闭源制造出来的技术神秘感，同样也能看出现在 AI 创业有多艰难。。

[lifcc @mylifcc]

step-5在鹈鹕跑分还可以啊

[SuSu_酥酥👅 @NFT_Chen]

🤨笑死！4月刚注册、注册资本20万的太原承明科技，经营范围还写着“人工智能基础软件开发”，成立5个月就能甩出一份12页法律函，指控ZCode偷偷打包上传了32932个文件、4.11亿明文字符，里面有完整源码、Git历史、数据库口令、API密钥和云凭证。

兄弟，你这5个月是把整个宇宙都写进仓库了吗？

20万注册资本养出4亿字符商业秘密，这效率比大模型训练还猛���。

建议智谱先别急着删数据，先问问这公司是不是用ZCode自己写的那份12页函。

#ZCode #智谱 #数据安全 #AI编程工具 #太原承明科技

[Yugui Yu @YuYuguiLab]

【B站上线“AI 无限竞技场”LLM 测评榜：号称“全球百大模型同场竞技”、GPT-6 现居榜首】首轮榜单显示 GPT-6 Astra 暂居榜首，前五名中国产大模型占三席；测评由 UP 主自主命题，还原真实场景表现，排名实时更新。https://t.co/yfyH5K0LlS

[Yugui Yu @YuYuguiLab]

@StepFun_ai @arena test it

[StepFun @StepFun_ai]

Step 5 Preview is built for professional knowledge work — from large-scale research to structured analysis and interactive reporting.

It coordinates research at scale and turns evidence into finished, auditable deliverables. In one agent action, it coordinated 950 web fetches and assembled 300,000 monthly records across 1,000 locations over 25 years; in another, it produced a 17-sheet analytical workbook with source reconciliation, formulas, and trend models.

Its outputs span technical engineering, creative production, analytical reporting, and public communication.

[StepFun @StepFun_ai]

Step 5 Preview works across software environments and sustains execution over long horizons.

Its capabilities extend from software engineering and web applications to 3D workflows and programmable hardware.

Over longer horizons, Step 5 Preview keeps track of prior results, uses execution feedback to decide what to try next, and continues iterating. We test this behavior in runs lasting up to 24 hours, including tasks involving GPU kernel optimization and automated post-training.

[Yugui Yu @YuYuguiLab]

@ChineseWSJ 最后一句话逻辑反转了！华尔街日报居然也是个草台班子😄

[StepFun @StepFun_ai]

Introducing Step 5 Preview: Advancing the Pareto Frontier.

Step 5 Preview is our new flagship model for agentic work, delivering frontier-level performance across software engineering and professional knowledge work, with particular strength in finance.

- 600B total / 27B active MoE, with 1M context + Vision
- Substantially lower task cost at comparable intelligence
- Broad software engineering capabilities with sustained execution over long horizons

Try Step 5 Preview: https://t.co/fC7HHlWKFn
Model page: https://t.co/4caJR2YGD3

Open weights on Oct 15.

[Qwen Developers @QwenDevs]

Qwen-Image-2.1 open source in 10hrs