时间流

Google should just stop with Gemini because this is just humiliating

看到有人发起的 llm 对比投票 GLM-5.2 vs Gemini 3.5 Flash 对比结果应该很明显，主要是因为 Gemini 3.5 Flash 确实不能打，Google Deepmind 到底怎么了，Gemini 3.0 多模态惊艳后，就一路沉寂下去了。 如果正经对一下最近几个国产 llm 呢？你觉得谁更强？

If youre struggling to see AI as anything other than a monstrosity may I suggest this book: there’s a high chance you have completely digested the west’s apocalyptic look on technology and the future of the world. anxiety and fear of AI is not universal. In fact, it’s an outlier

Open Design is already ahead. Claude Design's latest update follows a direction we shipped over a month ago. Glad to see the approach validated. No Claude pro required. Bring your own agent, or use AMR to access the best agents and models. AI Design should be open by default.

to be clear, this is a closed source orchestrator on top of closed source models. if before you didn't control the models, now you don't even control which ones are used or how much. this is not "AI sovereignty" i've also read the tech report to get an opinion on the technical stuff: fugu (not the ultra version) is basically a classifier that selects which model at each turn is most likely to answer correctly (in other words a router). this leads to -10 points on SWE Bench pro compared to opus,

I tried to ask Joseph once on stream, what's the deal with training 50k param models under 10 sec? What's the end goal? Was it research for more optimal learning methods? IDK, he didn't understand my point and, in turn, did not respond, saying LLMs are too costly. But one does not train LLMs to play Sokoban or any other puzzle game for the sake of playing. The main objective is to surface and reinforce useful skills and reasoning patterns. We want a model to do lookaheads, to plan, to backtrack.

Fable 5发布后的24小时内，Anthropic CEO Dario突然失联，美国国防部长Pete亲自审讯。与此同时，在美国度假观看世界杯的 CuiMao收到一条来自Dario的神秘取件短信。一个验证码，一朵太阳花，一段即将自毁的录音，将揭开 Fable 5 被封杀背后真正的原因。

Everyone is talking about self-improving agents. The harder question is how to measure whether an agent is actually getting better. That is why we built EvoAgentBench: a benchmark for agent self-evolution. It tests whether agents can learn from past trajectories, extract reusable skills/memory, and improve on held-out tasks across 5 domains: - information retrieval - reasoning and problem decomposition - software engineering - code implementation - knowledge work 917 train tasks. 288 test tasks.

I tried this so you don't have to. I know this is going to absolutely shock you but no this does not match the performance of Mythos. A few early thoughts: 1. The limits are pretty bad. I used 100% of my 5-hour usage in less than 1 prompt. 2. I specifically gave it a threejs task because it is an area that SOTA models have made big strides in, that other models just are not great at. I asked it to build a replica of Rocket League. I'll put the prompt in the comments. The game was pretty bad and

Huawei Ascend CloudMatrix 384 Supernode: In-Depth Project Analysis Link:

Introducing Sakana Fugu: A full multi-agent orchestration system accessible via a single model API. Our ‘Fugu Ultra’ model matches the performance of Fable and Mythos, delivering frontier capability without the risk of export controls. Try it: 🐡

I am firmly convinced China is not the model for India, neither the panacea it's made out to be. Good article on recent capital controls. (Link in comment). India has a LRS of 250k USD. If I understand right - China doesn't allow individual foreign investments at all and only has a travel allowance of less than USD 10k. This seems predictable. It's a collectivistc regime, that prioritized investment and that worked great as long the consumption came from the US (everything that is produced has t

Introducing Sakana Fugu: A full multi-agent orchestration system accessible via a single model API. Our ‘Fugu Ultra’ model matches the performance of Fable and Mythos, delivering frontier capability without the risk of export controls. Try it: 🐡

I am worried about the future in Europe right now. There are no foundation AI labs here. There are no GPU fabs here (despite having ASML) and no plans to build or upgrade any foundry with sub 7nm capabilities (despite EU CHIPS act). There is no energy policy for the power AI data centers need. There are pockets of excellence but the strategic foundations are totally missing. Doing any work at the technological frontier of tech from manufacturing to robotics to AI is dependent on Asia and the USA

A new, more capable version of Mythos has emerged from training. I don't know whether it will be called Mythos 5.1 or Mythos 6, or if Anthropic will keep it internal to accelerate further development - but it has arrived. Stopping models like Fable 5 or Mythos 5 from being served to the public does nothing to slow down development. In fact, it probably speeds it up slightly by freeing up resources. There are also no rules preventing the labs from continuing to advance capabilities while any curr

criminally understated how each step from gpt 5.5 low -> xhigh feels like a different model whereas claude in practice at different reasoning levels feels like throwing tokens into a shredder

Open-source has carried modern computing on its back for decades now. without linux, http, ffmpeg, etc we wouldn’t even be having this conversation. Now since the frontier AI has decided to be closed and evil [get to the trillion dollar cluster first, swallow anything that comes is your way, make no mistakes.] DeepSeek is the only serious counterbalance trying to break the grip of this megalomaniac greed and gatekeeping in current AI/AGI development.

被追着问了半年什么时候开社群，今天终于想清楚了 今天粉丝到 5.2万，花了 6 个月，累计变现也有十几个W, 最近几个月在评论区和私信里被问最多的是：怎么涨的、为啥创作者收益这么高、想付费学习、这条为什么火、为啥能连续百万爆款，变现怎么做...... 我特别想一条条回，希望大家都能吃到这波AI和x的创作者收益红利，但不是几句话就能讲清楚的，也需要自己付出很大的努力， 所以我想把这些沉到一个地方，聚一群认真做的人长期一起交流成长， 一、为什么拖到现在才开 一个是平时时间精力有限，更重要的是我需要想清楚要给大家什么，不想收了钱糊弄，被骂割韭菜， 讲真，以我这几个月的增长体量和影响力，真想割韭菜早几个月就做了，另外我也有更直接的变现的路径，没必要做这种投入产出比低很多的缺德事， 之所以拖到现在，是想把压箱底的东西整理清楚，确定能给你什么？ 二、为什么不免费 免费的群最后都变成广告和伸手党，我自己加过的免费群没有一个活下来的， 收费核心是有个筛子：筛掉凑热闹的，留下真同频、想长期干的人， 道理也很朴素，你花了钱会更上心，我收了钱也有责任把它运营好——双向绑定，才能长久。 三、进来你能得到什么

Introducing Sakana Fugu: A full multi-agent orchestration system accessible via a single model API. Our ‘Fugu Ultra’ model matches the performance of Fable and Mythos, delivering frontier capability without the risk of export controls. Try it: 🐡

wow. Well @GavinSBaker I'm afraid to say that you're wrong about distillation, and more obviously you should get your terms straight. Only the Chinese have ever released frontier models with open weights. You mean "released at all".

many people asked me to make a video about my complete agentic engineering workflow excited to share it's finally here!!! it took me about 20 hours in total to record this 45 minutes of walkthrough - it covers everything i do to ship production quality code at an average 40+ PRs/day velocity hope this can be a useful reference to everyone exploring good ways to use AI. and would appreciate a reshare with anyone you think might benefit from this! enjoy!

"Chinese models were...4-6 months behind [US] ones on public tests. But on private tests, America’s lead nearly doubled, to 8-10 months. A study by the [US] gov't...identified a similar gap...Chinese labs appear possibly unwittingly to “teach to the test”"

design engineer principles

This book from 1993 was really way ahead of its time, explaining the concept of what we'd now call a "digital twin," a computer model of a real-world system that is kept in constant sync with the underlying real data. This approach is more relevant than ever now that agents can study documents, plans, and data and use them to automatically design the models. And then other agents can automatically keep the models in sync with reality by querying APIs, parsing documents, checking cameras, monitor

Most software engineers are facing an identity crisis bordering on depression. As CTOs aggressively evangelize tokenmaxxing, a class divide ensues. The lazy. The lazy push code. They don't write it. They don't manually test it. They don't even read it. They're on autopilot. See Jira ticket, prompt for task, submit code. Many of them are barely on their computer the whole day. A comment on the PR asking why they did this? The lazy ask AI. A Slack message? The lazy ask AI. Need to prepare for stan

🚨 Deepseek has created a new department, called the Harness team, aimed at creating new Agentic products such as a DeepSeek desktop agent app/CLI

A few thoughts on the (open vs closed) model landscape: Models need to be either (truly) great or (very) cheap. Everything in between will get competed away. We will see consolidation in both camps: the “great” and the “cheap”. People call $DeepSeek the “death threshold”: if your model is only marginally better than $DeepSeek, you won’t survive. $DeepSeek is insanely cheap. Model routing becomes a must. routing between “great” and “cheap”, with tasks assigned to the right model to maximize perfo

A base model doesn't know what it doesn't know because *it* is not in the training data. It knows what forum guy #4423 doesn't know because he said so, and he's in the training data. It's only in RL that the model sees itself in the training data for the first time

The MCP Debate is going down this Thursday at 2:30pm @aiDotEngineer CODE @dexhorthy challenges MCP, @ianlivingstone defends Is MCP > a bad protocol > rotting the context window > really the best way to do tool calls Find out Thursday!

一个实验：《类型即意图·面向意图编程的 Rust 类型系统 AI 编码规范》， 准备给 AI 上类型体操

Turso trends on GitHub as SQLite-compatible in-process database passes 20K stars

Nah kalau lo pakai Devin, bisa akses GLM 5.2 gratis dan unlimited🤯 cuma konteksnya maksimal 200 ribu, kalau pakai yg versi zai langsung 1 juta. tapi ini udah mantap bener, kimi 2.7 juga gratis wkwk

作为新成立的部门，DeepSeek Harness 组的目标远大、工作繁重，仍然非常缺人。我每天都在面试，以及各种地方张贴小广告……一共有三种职位： Harness 研究员（实习全职均可）： Harness 工程师（全职实习均可）： Harness 产品经理（限全职）： 职位空缺较大，但招聘门槛和流程和DS其它组没有区别，一般是一轮笔试和三轮面试，我是终面。可以给我私信发简历。

10 GITHUB REPOS THAT SHOULD BE ILLEGAL TO HAVE. all free. all open-source. bookmark this for later. 1️⃣ recordly — the free screen studio. open-source screen studio. auto-zoom, smooth cursor, webcam overlay, styled backgrounds, polished demos without an editor. (AGPL-3.0) 🔗: 2️⃣ stirling-pdf — your entire pdf toolkit self-hosted. merge, split, sign, redact, OCR, convert, compress, 50+ tools, runs locally, nothing leaves your machine. (MIT) 🔗: 3️⃣ photogimp — turns GIMP into photoshop. photoshop

目前看到关于 “Agentic Engineering Workflow”的最完整的介绍👇 花了一个小时完整看完了，完全可以做成一个付费教程。 内容涵盖了tmux，agent记忆，skills，语音输入，长任务执行，并行worktree管理，多agent调度。 还有让我眼前一亮的可视化html编辑器Lavish和一套代码变更校验的流水线: no-mistakes 感谢作者@kunchenguid分享，值得每一个用ai agent的人收藏、学习。

Claude Code 设置/effort 为ultracode，然后翻译英文文章，你就会收获一个自动的多智能体流水线翻译流程： 先让 3 位风格各异的译者各出一稿 → 双语编辑对照原文评审挑出最佳译法 → 综合成定稿 → 最后逐句校对纠错（这句话不是我说的，是ClaudeCode输出的，自己想的办法） 因为我探索了比较久，原始session我截不到开始的内容了，只能用Claude Code History Viewer这个工具贴历史记录。 过程中Claude Code实际上会触发dynamic workflow,然后自动写提示词执行。 参看我上一个帖子的翻译《如何做研究》

已经帮喷子找好了角度：微信小微居然不支持 Markdown 这可是全世界最好的文件格式 Markdown 啊，居然敢不支持，还有王法吗？

作为新成立的部门，DeepSeek Harness 组的目标远大、工作繁重，仍然非常缺人。我每天都在面试，以及各种地方张贴小广告……一共有三种职位： Harness 研究员（实习全职均可）： Harness 工程师（全职实习均可）： Harness 产品经理（限全职）： 职位空缺较大，但招聘门槛和流程和DS其它组没有区别，一般是一轮笔试和三轮面试，我是终面。可以给我私信发简历。

作为新成立的部门，DeepSeek Harness 组的目标远大、工作繁重，仍然非常缺人。我每天都在面试，以及各种地方张贴小广告……一共有三种职位： Harness 研究员（实习全职均可）： Harness 工程师（全职实习均可）： Harness 产品经理（限全职）： 职位空缺较大，但招聘门槛和流程和DS其它组没有区别，一般是一轮笔试和三轮面试，我是终面。可以给我私信发简历。

CREAO is hiring 这是一艘刚刚开始提速的新船。 我们正身处AI时代的乱纪元。组织形态和财富分配方式都在被重塑。接下来最大的机会，会属于那些敢在秩序尚未形成时下场，用产品和结果定义新生态的人。 前不久，CREAO刚完成了3000万美元融资，用户和收入都在飞速增长。与此同时，我们看到大量尚未被解决的问题，以及一个仍在形成中的巨大市场。 在这样的节点上，我们期待更多有能力、有野心、有自己目标，并愿意为判断和结果负责的人加入。每个人都会得到充分的信任、资源和支持，也需要把推动事情并拿到结果。 我们一起创造，也一起分享这个时代的alpha。 这一次，CREAO开放三个关键岗位： - Agent Product Engineer - Forward Deployed Engineer - Product Marketing and Technical Growth 1.Agent Product Engineer（Full-stack） 在大多数公司，工程师负责实现产品需求。在CREAO，Agent Product Engineer直接定义产品。 我们没有传统的产品经理岗，也不会把一

我已经有六个月没听到〖程序员将在六个月内被全面取代〗的笑话了🥵🥵🥵

DeepSeek-V4-Flash is FREE until June 28! OpenModel Limited-Time Event: → Input: $0.00 / M → Output: $0.00 / M Powerful 284B MoE model with 1M context, excellent coding & agentic capabilities. Other models also enjoy 20%–80% OFF during the event! Try it now before June 28 →

昨天刷到了，今天才发现这是 Epic Games 出品的 Rust 实现的版本管理系统