Luke Wang

[少楠Plidezus @plidezus]

分享下 flomo 认知地图的设计思路。

1. 把笔记向量化（本来相关笔记也会用到），然后压缩到 2D 获得坐标；
2. 根据坐标获得等高线，绘制出来 2D 认知地图
3. 再用 @threejs 还原成 3D效果，可以近距离观察。
4. 加上笔记创建的时间轴，就知道自己记录的山峰是在什么时候攀登的。

[Ning @totheagi]

we got the full Kimi K3, 2.8T params, running on 80x RTX 5090s.

20 tok/s single stream, day one, untuned. Last week we took GLM-5.2 from 30 to 110 tok/s on this same fleet. This number will climb.

A first for open weights: frontier intelligence served with zero HBM, the scarcest silicon in AI. Just GDDR7 gaming cards, plain ethernet, and the official MXFP4 weights, nothing requantized.

The most powerful open model on Earth, on the most abundant GPUs on Earth. Any lab, startup, or university can now own it, probe it, fine-tune it, run agents on it.

@Kimi_Moonshot

[Jakub Antalik @Jakubantalik]

Thinking Orbs, animated component library

6 types, optimized for large and small sizes, light & dark mode, playground, zero dependencies

https://t.co/PdGxwi6Kws
npm install thinking-orbs

Collaborated on this with @a_brinza who did a lot of work

[Dilum Sanjaya @DilumSanjaya]

Fun interactive science app ideas | Part 10

Built an app for exploring different types of butterflies

It's surprising how calm and peaceful the gentle wing movements feel

Everything is generated with code. No 3D models

Code
Kimi K3

[Joshua Guo @jshguo]

Created a smooth globe-to-map transform. Made with @v0.

If the geography teacher taught you with this animation, would you still hate geography?

[Qwen @Alibaba_Qwen]

🚀 Introducing Qwen3-Omni — the first natively end-to-end omni-modal AI unifying text, image, audio & video in one model — no modality trade-offs!

🏆 SOTA on 22/36 audio & AV benchmarks
🌍 119L text / 19L speech in / 10L speech out
⚡ 211ms latency | 🎧 30-min audio understanding
🎨 Fully customizable via system prompts
🔗 Built-in tool calling
🎤 Open-source Captioner model (low-hallucination!)

🌟 What’s Open-Sourced?
We’ve open-sourced Qwen3-Omni-30B-A3B-Instruct, Qwen3-Omni-30B-A3B-Thinking, and Qwen3-Omni-30B-A3B-Captioner, to empower developers to explore a variety of applications from instruction-following to creative tasks.

Try it now 👇
💬 Qwen Chat: https://t.co/rJ5DUlBJBr
💻 GitHub: https://t.co/VHHp1jWcHT
🤗 HF Models: https://t.co/WsP8vyWltp
🤖 MS Models:
https://t.co/sGnbZRfxVz
🎬 Demo: https://t.co/v2WrZvCKD6

[歸藏(guizang.ai) @op7418]

这个也是考虑复杂的时序提示词和变化的测试，图片本身只有一个穿着便服的女孩，她的服装需要顺畅的在后面几秒不断变换。

可以看到处理非常顺滑，公主长裙和披风出现的时候是用一种生长动画的方式，然后宇航服出来的时候又是类似蒙版的方式，整个过程跟原来的人像一致性非常高，跟周围环境的交互也很自然。

[-Zho- @ZHO_ZHO_ZHO]

nano-banana 这恐怖的数据卧槽，比当时 GPT 4o 还要猛

好模型就是狠，Google 翻身仗打得漂亮

单日破百万浏览量，差不多等于之前 GPT 4o 5 天的量了

[howie.serious @howie_serious]

chatgpt最新咒语：quizgpt 🪄

除了study mode，chatgpt 还偷偷上线了（天然内置ai所以秒杀传统闪卡的） flashcard 功能！

使用这条咒语，你可以立刻 quiz 一切：

> quiz me with QuizGPT：{主题，或任何prompt}

例如，quiz LLM基本概念，quiz 《春江花月夜》……一切知识，一切内容。

甚至，你可以quiz你和chatgpt刚讨论过的话。或者在prompt下面复制粘贴你要quiz的文本。考虑到gpt-5巨大的context window，使用场景巨大。

quizgpt 是openai对这个功能的代号，使用quizgpt这个关键词就可以激活。偶尔失败的话，可以直接指明工具名称，像这样：

> quiz me with QuizGPT（ecosystem_demo.flashcards）

每次测试，chatgpt会自动生成难度从低到高的多套flashcard，先是基础，然后是进阶，然后是大师，然后是宗师……

chatgpt flashcard（quizgpt），因为内置于最顶级的llm，和之前的flashcard工具在本质上是完全不同的。我玩了好久，感觉可以成为一个日常使用频率很高的利器。大家赶紧测试起来~~

[Massimo @Rainmaker1973]

The first direct observation of gravitational waves captured the merging of two black holes which emitted 36 septillion yottawatts of power (3.6×10⁴⁹ watts), greater than the combined power of all light radiated by all the stars in the observable universe.

[🎞️ SXS simulation]

[dangjin @JinsFavorites]

GPT4o - 教育场景

一句话生成单词闪卡

[Jeremy Nguyen ✍🏼 🚢 @JeremyNguyenPhD]

ChatGPT's new Image Generation dropped less than 24 hours ago

Here are 15 great examples of what you can do now, some limitations—and a hidden trick to get instant access if you're still waiting!

1. Life-like photos:

[Luke Wang @Remember2015]

@ripe18 效果相当不错，点赞

[Dreaming Tulpa 🥓👑 @dreamingtulpa]

CtrLoRA can adapt a base ControlNet for image generation with just 1,000 data pairs in under one hour of training on a single GPU!

It reduces learnable parameters by 90%, making it much easier to create new guidance conditions.

https://t.co/oBnh7WR2rH

[Luke Wang @Remember2015]

@dreamingtulpa only SD 1.5?

[Weizhu Chen @WeizhuChen]

We updated Phi-3 mini in our June release, with the enhancement in instruction following, reasoning in MMLU (70.9)/GPQA(30.6), and better long context. Share with us your feedback on the new models
https://t.co/Yg4Awrvu1K
https://t.co/4UIpTHnZ9P

[Luke Wang @Remember2015]

@liujinmarshall 为啥还要交处理费？

[Jerry Liu @jerryjliu0]

Here’s a neat paper by Barnett et al. (@DeakinA2I2) that outlines 7 failure points in building a RAG pipeline over your data.
🚫 Missing content (did not index it)
🚫 Missing in top-k retrieved set
🚫 Missing in reranked set
🚫 Not extracted (in context but LLM couldn’t use)
🚫 Wrong format (e.g. JSON)
🚫 Incorrect specificity (not at the right level of granularity)
🚫 Incomplete - the synthesized answer only answers part of the question

We've posted a lot about this on the @llama_index side but this diagram nicely covers a lot of the aspects. If I were to add a few, I’d add failure points during the query understanding/rewriting phase (particularly if you’re building agents).

Check out the paper: https://t.co/PPH9PLq3gj

[OpenMoon @OpenMoonAI]

This guy shoots a frozen lake and the bullet spins - Inception style.

[iPaul @iPaulCanada]

在经济通缩的大环境下，取消艺术特长生以后，钢琴市场一片惨淡。