Ethan Sun (@SunUpali) header
Ethan Sun (@SunUpali) profile picture

🚀AI · 出海 · 思考者 | 用Claude写代码,用感悟治焦虑 |同好互关,共同进步🤝 Ethan Sun | Road not taken, but turning anyway.

Joined April 2012
Photo by Ethan Sun (@SunUpali)Photo by Ethan Sun (@SunUpali)Photo by Ethan Sun (@SunUpali)Photo by Ethan Sun (@SunUpali)Photo by Ethan Sun (@SunUpali)Photo by Ethan Sun (@SunUpali)Photo by Ethan Sun (@SunUpali)Photo by Ethan Sun (@SunUpali)Photo by Ethan Sun (@SunUpali)Photo by Ethan Sun (@SunUpali)
Ethan Sun retweeted
“穿上它!你紫腚行!” 岁月是把指甲刀,时光温柔如内裤。 纯棉日抛内裤,霸总恋爱必备。 守护女性幻想,避免自己暴露在互联网的风险当中,成为段子的主角。 女性送礼佳品, 爱他,就给他蛋蛋的陪伴。 内裤文学鼻祖、颖学创始人亲笔签名,30天【预售】发货。 鸡飞蛋打的爱情, 一举两得的守护。 纯棉呵护的温柔, 一周七抛的奢华。 不做随便男孩, 从穿全棉日抛内裤开始。 利润部分将捐赠给流浪动物保护组织,用于中国境内流浪狗救助公益事业。 【淘宝】e.tb.cn/h.8Px7mHXXC4q1MdG?tk… tG-#22>lD 「亲笔签名男士纯棉日抛内裤七条装护蛋不求人」 点击链接直接打开 或者 淘宝搜索直接打开
632
183
77
1,933
4,023,428
Ethan Sun retweeted
Grok Build now lets you resume your recent Claude Code, Codex, and Cursor sessions directly from the main prompt Just type: • /resume-claude • /resume-codex • /resume-cursor You can instantly pick up a recent coding session and continue working with Grok 4.5 inside Grok Build No need to start over, rebuild the entire context, or explain everything you were working on Grok picks up where you left off and lets you continue your workflow seamlessly
47
65
10
472
21,130
Ethan Sun retweeted
As engineering, product, design, DS, etc. melt into a new kind of role, I was reflecting on what roles might look like in the future. For example, when I look at the Claude Code team I see what I think is five archetypes: 1. Prototyper: comes up with brand new ideas; churns out many ideas, most of which don't ship 2. Builder: quickly turns a prototype/idea into production-grade product/infra 3. Sweeper: cleans up the UI, simplifies the code and system, unships, optimizes performance 4. Grower: takes a product that has been built and iterates on it to improve Product-Market Fit 5. Maintainer: owns a mature system to make it secure, reliable, fast, and efficient as it scales Many people span across 2 roles, and sometimes 3 roles. I also notice that these roles are not really tied to job function -- eg. across Anthropic, some designers match category 1, some 2, some 3; same for engineers, PM, DS. A healthy team needs a mix of these, depending on the product: - A product that is new and pre-PMF needs people that are strong at 1+2+3 - A product that is growing and has found PMF needs 2+3+4 and some 5 - A product that has strong PMF needs 3+4+5 and some 2 Maybe product roles of the future will look more like this, and less like the domain-specific roles of today?
883
2,457
694
20,130
3,184,017
Ethan Sun retweeted
Anthropic’s complaint has all the hallmarks of an incumbent technological leader unsettled by the prospect of capable rivals and eager to enlist political power to preserve a market advantage. Technological diffusion is not a Chinese invention. It is how economic history works. Europe borrowed silk, porcelain, papermaking, and countless other technologies from China. The United States built much of its early industrial strength by appropriating British manufacturing know-how. Every great technological power has learned from its rivals. Yet when Chinese firms become capable competitors, we are suddenly told that the normal circulation of knowledge is an existential threat. If Anthropic’s capabilities can genuinely be replicated through question-and-answer interactions, that is less an indictment of Chinese ingenuity than a confession about the fragility of Anthropic’s own competitive moat. A company confident in its technological lead does not rush to Washington every time a rival catches up. Beneath the rhetoric of “AI theft” lies a simpler anxiety: Chinese AI firms are improving rapidly, and American firms fear losing market dominance. The real issue is not espionage but competition. History offers a clear lesson. Technological leadership is never preserved by complaining that others are learning. It is preserved by innovating faster than they do. Anthropic appears increasingly interested in the former because it is less certain of the latter.
41
90
9
533
67,308
Ethan Sun retweeted
How did GLM-5.2 (Max) get to the top of Code Arena: Frontend? Looking at matched head-to-head on real-world web dev frontend tasks, @Zai_org's latest model takes a higher win share than its opponent in every pairing but one. - Beats every Claude Opus variant head-to-head: including Opus 4.8 (Thinking) and Opus 4.7 (Thinking) - Widest margins come against strong models: 61.0% vs Kimi-K2.6, 59.4% vs Sonnet 4.6 and 55.0% vs Opus 4.7 (Thinking). - The closest fight isn't from the very top: GPT-5.5 (xHigh), ranked #16, pushes it hardest at 41.7% to 40.0%. Opus 4.6 is the other (47.0% to 42.4%). There is a generational nuance here. Versus its own predecessor GLM-5.1, the latest model is tied: 45.5% - 45.5%. The jump shows up against the field, the one pairing without a clear win is in a direct rematch with 5.1.
Exciting news: GLM-5.2 (Max) ranks #2 in Code Arena: Frontend, with +29pt over Claude Opus 4.7 (Thinking) and only behind Fable 5! GLM-5.2 is the best open model vs Kimi-K2.6 and Minimax-M3 by a large margin. - #2 React and #4 HTML sub-leaderboards - Ranks as the top model in nearly all sub categories: Brand & Marketing, Reference-Based Design, Data & Analytics, Consumer Product, Gaming, and Simulations. Congrats @Zai_org for the incredible milestone!
32
43
10
556
89,407
Everyone benchmarks GLM-5.2 against the frontier now. So we did too. We pulled GLM-5.2's plan up against Claude Fable 5's, the plan that won our last frontier round. Same prompt, same task, same rubric. Fable scored 9.1. GLM-5.2 scored 9.0.
100
166
58
2,457
243,480
Ethan Sun retweeted
上班有一条底线: 当你意识到自己升不上去了,工资也基本不可能再涨了 现在你最重要做的事: 锻炼身体,照顾家庭,发展副业。把尿喝白,把电充绿,把心放松。
138
100
5
917
54,213
Ethan Sun retweeted
I'm 55. If you're in your 30s or 40s, read this: 1. Marriage won’t fix loneliness. Neither will kids, money, or sex.
485
3,671
177
25,166
6,338,964
Ethan Sun retweeted
Dario is wrong. He knows absolutely nothing about the effects of technological revolutions on the labor market. Don't listen to him, Sam, Yoshua, Geoff, or me on this topic. Listen to economists who have spent their career studying this, like @Ph_Aghion , @erikbryn , @DAcemogluMIT , @amcafee , @davidautor
Anthropic CEO Dario Amodei: “50% of all tech jobs, entry-level lawyers, consultants, and finance professionals will be completely wiped out within 1–5 years.”
1,184
2,683
491
20,967
4,117,071
Ethan Sun retweeted
如何判断一个人是不是 AI Native? 很简单,交给他/她一个陌生任务。 一个传统的人面对陌生任务 —— 比如让文科生去写一段爬虫 —— 第一反应是古典时代的肌肉记忆:这不是我的专业,我不会代码,做不了。 一个 AI Native 的人,脑回路完全不同。 他不在乎自己会不会那门手艺。他只问三个问题: - 我能精确描述我想要达成什么结果吗? - 我能把这个大问题拆解成 AI 能理解的子任务吗? - AI 吐出来的东西,我能判断对不对吗? 三个都是“能”,他就能做。
50
122
16
711
130,611
Ethan Sun retweeted
新研发了一个 Web-To-Markdown Skill 不需要任何配置、不需要任何登录态,安装即用,支持 YouTube / 微信 / 知乎 / 飞书 / X / 等任意网页。 全部一键抓取清洗为 Markdown(Youtube 是自动拿字幕)。 简单、高效、极致,是我做 Skill 的习惯, 不改写、不洗稿、不加 AI 腔,只做一件事:输出可以直接喂给 Agent 的正文。 工作流就三步: 贴 URL → web-to-markdown → 直接用于翻译 / 摘要 / 知识库 安装命令: ```bash npx skills add rookie-ricardo/erduo-skills --skill web-to-markdown ``` GitHub: github.com/rookie-ricardo/er…
32
110
9
592
72,612
Ethan Sun retweeted
Zhipu launched AutoClaw today. Stock jumped 13%. That makes the list: Moonshot's KimiClaw, MiniMax's MaxClaw, Alibaba's CoPaw, ByteDance's ArkClaw, Tencent's WorkBuddy, and now Zhipu's AutoClaw. Every major Chinese AI company has built a product around one open-source project in a matter of weeks. This is not herd behavior. Each company has a specific structural reason to be here. I explained why in my latest analysis:hellochinatech.com/p/opencla…
37
154
30
1,009
148,204
Ethan Sun retweeted
如何判断自己是不是NPC?1、经常无聊,没有主线任务。2、不怎么消耗资源,没有钱。3、建模简陋,长得普通。4、只有逢年过节,才会更新自己的皮肤。5、生命值较低,容易受伤。6、活动受限,经常锁定在单一区域,去哪里都有约束。以上全部符合了,就基本可以确定了。
37
78
10
820
113,624
math is beautiful
220
2,183
254
34,660
1,009,668
Ethan Sun retweeted
The minute I become too frugal or money-conscious is the moment it starts to control me and becomes difficult to make. When I spend almost carelessly, it seems to find its way back. Money-induced stress narrows the mind and new opportunities become invisible.
423
594
99
6,644
365,939
Ethan Sun retweeted
一切以Claude Code为核心打磨工作流,Obsidian作为存储和查看器,语音输入优先。 核心工具是Skill、Cli和MCP,语音输入用闪电说,硬件用大疆mic3: ① 听歌:Spotify MCP + 音乐风格查找 Skill,语音输入播放任意歌曲。 ② 阅读:一句话下载电子书 + 读书笔记 Skill,全网AI Newsletter采集阅读Skill。 ③ 编程:这个不用多说。 ④ 写作:文章配图、生成封面、X/公众号发布,准备对应Skill处理。 基本就够了,持续打磨、迭代优化即可。
12
54
1
296
27,944
Ethan Sun retweeted
Anthropic 刚发了一篇报告,点名 DeepSeek、Moonshot(Kimi)、MiniMax 三家中国 AI 公司大规模蒸馏 Claude 规模:2.4 万个虚假账号,1600 万次对话。 其中 MiniMax 一家就贡献了 1300 万次 蒸馏是什么意思? 简单说就是用一个强模型的输出去训练一个弱模型。 你让 Claude 回答大量精心设计的问题,把问答对收集起来,拿去训自己的模型。 相当于让 Claude 当免费老师,教会你的模型怎么思考 这个技术本身是合法的,各家 AI 公司都用蒸馏来做自己的小模型。 但拿别人的模型偷偷蒸馏,违反使用条款,并且绕过了出口管制 三家的操作手法差不多: 通过代理服务商批量注册账号,用 proxy 网络分散流量避免被检测。 有一个代理网络同时管理超过 2 万个虚假账号,把蒸馏流量混在正常用户请求里 具体来看每家在偷什么: DeepSeek — 15 万次对话,主要提取推理能力。 有个操作很有意思,他们让 Claude 想象一个已完成回答背后的内部推理过程并逐步写出来,相当于批量生成 chain-of-thought 训练数据。 还让 Claude 生成政治敏感问题的"安全替代回答",大概率是拿去训自己模型的审查能力 Moonshot — 340 万次对话,目标是 agentic reasoning、工具调用、编程、计算机视觉。用了几百个账号走不同渠道,后期还尝试提取和重建 Claude 的推理链路 MiniMax — 1300 万次对话,最大规模。专注 agentic coding 和工具编排。Anthropic 在 MiniMax 还没发布新模型的时候就抓到了,完整看到了从数据生成到模型发布的全过程。并且 Anthropic 发新模型的时候,MiniMax 24 小时内就把近一半流量转向新模型 Anthropic 怎么发现的?通过 IP 关联、请求元数据、基础设施指标,有些直接追溯到了实验室具体的研究人员 这件事跟我们之前聊的 API 中转站数据风险是一个逻辑。 中转站坐在你和模型中间能看到所有数据,这些蒸馏攻击用的就是类似的代理架构,只不过规模大得多 Anthropic 把这事上升到了国家安全层面,说蒸馏出来的模型没有安全护栏,可能被用于军事、情报、监控。并且他们认为这些攻击说明出口管制是有效的——正因为芯片受限没法自己从头训,才需要来蒸馏别人的模型 评论区有个角度也值得注意: 有人问 Anthropic 既然能通过元数据追溯到具体研究人员,那对普通用户的隐私意味着什么?
We’ve identified industrial-scale distillation attacks on our models by DeepSeek, Moonshot AI, and MiniMax. These labs created over 24,000 fraudulent accounts and generated over 16 million exchanges with Claude, extracting its capabilities to train and improve their own models.
12
2
1
22
15,001