I will make it for my son and my family.
Joined April 2021
- Tweets16.8K
- Following1.5K
- Followers1K
- Likes16K
Ξlpida retweeted
真没想到！写设计文档最容易翻车的，不是写不出来，而是写的人自己看不出盲点。
Google 开源的 elephant-goldfish-model（EGM）就是冲着这个来的：1 万多 Stars，Apache-2.0。
以前靠同一个人、同一段上下文又写又审，越看越觉得自己对。EGM 把角色拆开——大象带着项目上下文起草稿，金鱼用零上下文来挑刺。
核心能力：
大象 Agent 带着项目上下文做 brainstorm / draft
金鱼评审用全新 subagent、不带早先对话来找盲点
覆盖 brainstorm、draft、review、resolve、bootstrap、drift report
可装到 Claude Code / Codex / Antigravity
评审分多阶段：清晰度、代码契合、实现风险
你拍板架构取舍，批准规格后再让 agent 写代码
适合「功能需要设计决策」的场景，小修小补可跳过
设计文档写完就直接开写代码的，值得先把评审这一环接上。
github.com/google/elephant-g…
Ξlpida retweeted
9,000+ stars on Google's Elephant-Goldfish Model, and the idea fits in one line: the agent that wrote the spec shouldn't review it.
Elephant drafts with full project context. Goldfish reviewers see only the doc and its sources, no chat history. You approve.
Claude Code, Codex, Antigravity.
github.com/google/elephant-g…
Ξlpida retweeted
google/elephant-goldfish-model: Open-source skills for designing and reviewing software with AI agents github.com/google/elephant-g…
Ξlpida retweeted
GITHUB STAR MOVERS · NOV 3, 2026 (13/15)
elephant-goldfish-model | Design with AI, review with fresh eyes.
What it is: A tool that uses AI agents to draft software designs and then reviews them independently for blind spots, helping teams make better architectural decisions.
Why it's rising: Crossed 10k stars today. +1,422 yesterday, +1,274 today, 10.9k total, only 50 days old. Developers need better design workflows to catch flaws early and reduce rework.
github.com/google/elephant-g…
Follow me for the fastest-rising GitHub projects every day!
Ξlpida retweeted
🔥 Google just launched something that changes AI coding.
Most agents jump straight to code.
Google forces design first.
🐘 Elephant = full context, drafts the spec
🐟 Goldfish = zero memory, attack every weak spot
Only survivors get coded.
egm.dev
Ξlpida retweeted
Elephant-Goldfish Model > Compound Engineering.
I plan to switch using EGM in all my projects.
egm.dev
Ξlpida retweeted
Google open sourced the planning tool it says thousands of its engineers use before AI writes code.
Plus a free 4B Microsoft model that runs your computer, and a free AI agent that works on your Mac offline. Today's free AI, with the catch.
Free tools
• Google's Elephant-Goldfish Model (EGM) is a plugin for Claude Code, Codex and Antigravity. One agent writes the plan, fresh agents pick it apart, and you approve before any code is written. About 9.6K GitHub stars.
Free models
• Microsoft's Mage-VL-Cowork is a 4B open model that clicks around your screen or jumps to the terminal, whichever is faster. Microsoft says it scores 66% on OSWorld-Verified. The human score is 72.4%.
Run AI locally
• Local Code from the Draw Things team is now in the Mac App Store. It's a free, open source agent that makes slides, charts and code with models on your own Mac, so your files stay with you.
The theme today is AI that works like a careful teammate. It plans first, picks the fastest tool and keeps your data at home.
The catch: EGM isn't an official Google product and burns extra usage. Mage's 66% is Microsoft's own test, and you need a GPU to run it. Local Code needs at least a 16 GB Mac.
#AI #OpenSource
Ξlpida retweeted
[GitHub Repo] Elephant-Goldfish Model: Open-source skills for designing and reviewing software with AI agents
github.com/google/elephant-g…
#MachineLearning #ChatGPT
Ξlpida retweeted
The Elephant-Goldfish Model: Why the AI That Knows Least Reviews Best ift.tt/nDqtoAY #AI #MachineLearning #ArtificialIntelligence #NaturalLanguageProcessing #LLM #FutureOfAI
Ξlpida retweeted
Literally is crazy
Makes sm sense
The elephant is your LLM
The goldfish, knows abs nothing and can spot out anomalies in responses and data, etc....
Think about all the companies that need this is their review of AI work
Crazy
$EGM coded
GOOGLE ENGINEERS REVIEW THEIR CODE WITH A F*CKING GOLDFISH...
it's called the elephant-goldfish model. open source, 3.2K stars
→ the elephant: your main agent. knows the whole project, writes the design doc
→ the goldfish: fresh subagents with zero memory of your chat. they read the doc cold and catch everything the elephant "just knew"
→ you: pick the trade-offs, approve the spec, then it builds
skills for brainstorm, draft, review, resolve, plus a drift report that checks your docs against the code
works in claude code, codex and antigravity
does your agent review its own work?
Ξlpida retweeted
Check the accounts posting bout $EGM🧏
ニュース「Google、Elephant-Goldfish Modelプラグインを公開 ——AIが設計文書の作成と検証を分担」公開 gihyo.jp/article/2026/10/ele…
Ξlpida retweeted
google, muhendislerinin kod yazmadan once spec yazip test ettigi ajan yontemini acik kaynak yapmislar. elephant-goldfish modeli
- fil: tum repo baglamini bilen ajan. brainstorm yapiyor, spec'i yaziyor
- japon baligi: hafizasi sifir subagent'lar. konusmayi hic gormeden sadece spec'i ve kodu okuyup acik ariyor
- sen: trade off'a karar veriyorsun, spec'i onayliyorsun, ondan sonra kod.
uzun bir konusmada ajan da sen de ayni varsayimlara aliskanlik yapiyorsunuz. baglamsiz goz yazilmamis varsayimi yakaliyor.
icinde 7 skill var, benim en fazla ilgimi ceken drift-report. kod yazildiktan sonra spec'ten nerede saptigini cikariyor.
github.com/google/elephant-g…
Ξlpida retweeted
Ticker is $EGM
REAL TEK ON GOGGLE GITHUB THAT TRENCHES SLEEPING ON
Ξlpida retweeted
We are so early on $EGM
Soon others will see the vision
It’s open sourced by goggle and on their official repo, lotta goggle engineers use it everyday
It’s relevant real tech and trenches are sleeping on this
Google 开源了一个很有趣的 AI Coding 插件，叫做 Elephant-Goldfish（大象与金鱼）。
它想解决一个常见问题：
你跟 AI 讨论了两个小时，AI 觉得自己什么都懂，结果写出来的程序还是跟需求有落差。
Google 的做法是：让两个 AI 代理分工。
⸻
🐘 大象（Elephant）
负责记住项目脉络、讨论架构、整理需求，把决策写成设计文件。
🐟 金鱼（Goldfish）
刻意不继承之前的聊天记录，像第一次接手项目的工程师，重新审查设计。
金鱼会做三件事：
理解检查： 光看文件，能不能理解整个设计？
对抗审查： 有没有遗漏边界条件、隐藏依赖或逻辑问题？
实现检查： 文件、数据结构、函数接口是否清楚到可以开始写程序？
如果金鱼看不懂，就回去修改规格。
⸻
这套流程最有意思的地方，是把 AI Coding 的质量管控往前移。
以前是 AI 写完程序，人类才开始找问题。
现在可以在写第一行 Code 之前，就先让另一个 AI 挑出规格漏洞。
尤其适合跨文件修改、复杂功能、多人协作的项目。
目前官方 GitHub 已列出 Antigravity、Claude Code、Codex 的安装方式。
不过，审查通过仍不代表程序一定正确，人工验收与测试还是需要。
⸻
我觉得值得观察的是：
当 AI 越来越会写程序，「谁来检查 AI 理解得对不对」，可能会变成更重要的工作。
项目网站：github.com/google/elephant-g…
Ξlpida retweeted
4QV31UeJJyUVWP2f1qgEz8gUPkf6WyPZyAmGdT8UD8Ld
$OPCAT
Hello CT I’m OP_CAT 👋
OP_CAT was originally an opcode in Bitcoin. It was disabled in 2010 but slight variations on it are frequently proposed to be added to Bitcoin using a soft fork
Now I am back to LIFE 👀
4QV31UeJJyUVWP2f1qgEz8gUPkf6WyPZyAmGdT8UD8Ld
$OPCAT
