Ling

Have 2 cats | Big fan of Yoga | Work in Tech | Study CSS

[Josh Z @joshzjs]

AgentDrive private beta is live 🎉

It's the "Google Drive" for your agents!

A persistent file/workspace layer for AI agents - accessed primarily through MCP. Create, version, restore, upload, and share agent work with workspace-scoped authorization.

Try it: https://t.co/sYg4Gu5o74

Feedback welcome!

[Attio @attio]

Introducing GTM Atlas, a map for modern AI GTM built with some of the best operators in the industry.

A free resource covering the full customer journey, from lead capture to expansion, with the systems thinking that scales with you.

[Jacob Coxon @hilbertspaess]

I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.

[Matt Dailey @reactiverobot]

Here's my advice for engineers looking to up their design game. https://t.co/oBFur0Izaf

[Ling @xinxialynn]

@elliotchen 网页设计好有意思！首屏真的很像推理游戏开头

[Lun Wang @lunwang1996]

I’ve left Google DeepMind after an amazing chapter.

I’m incredibly grateful for the people I worked with, the things we built, and the lessons I learned from taking frontier AI research into production. DeepMind shaped how I think about research, product, evaluation, and what it takes to build AI systems at real scale.

As I wrap up this chapter, I wrote down something I’ve been thinking about a lot: evals.

We’re good at evaluating the models we have. We’re much worse at evaluating the models we’re about to build — especially if they cross into a new capability regime. We will have self-evolving models, but before that, we need self-evolving evaluations.
https://t.co/F1lUWxDG2D

[Ling @xinxialynn]

@YASLant @0xKevin00 @suannaigg credits ≠ token

[Ling @xinxialynn]

@ScottCao77 @h7J3d me too!!!!

[Anthropic @AnthropicAI]

Introducing Project Glasswing: an urgent initiative to help secure the world’s most critical software.

It’s powered by our newest frontier model, Claude Mythos Preview, which can find software vulnerabilities better than all but the most skilled humans.
https://t.co/NQ7IfEtYk7

[Ling @xinxialynn]

Amazing

[Elliot 艾略特 @ElliotChen]

论文来了。名字叫 MSA，Memory Sparse Attention。

一句话说清楚它是什么：
让大模型原生拥有超长记忆。不是外挂检索，不是暴力扩窗口，而是把「记忆」直接长进了注意力机制里，端到端训练。

过去的方案为什么不行？
RAG 的本质是「开卷考试」。模型自己不记东西，全靠现场翻笔记。翻得准不准要看检索质量，翻得快不快要看数据量。一旦信息分散在几十份文档里、需要跨文档推理，就抓瞎了。

线性注意力和 KV 缓存的本质是「压缩记忆」。记是记了，但越压越糊，长了就丢。

MSA 的思路完全不同：
→ 不压缩，不外挂，而是让模型学会「挑重点看」
核心是一种可扩展的稀疏注意力架构，复杂度是线性的。记忆量翻 10 倍，计算成本不会指数爆炸。
→ 模型知道「这段记忆来自哪、什么时候的」
用了一种叫 document-wise RoPE 的位置编码，让模型天然理解文档边界和时间顺序。
→ 碎片化的信息也能串起来推理
Memory Interleaving 机制，让模型能在散落各处的记忆片段之间做多跳推理。不是只找到一条相关记录，而是把线索串成链。

结果呢？
· 从 16K 扩到 1 亿 token，精度衰减不到 9%
· 4B 参数的 MSA 模型，在长上下文 benchmark 上打赢 235B 级别的顶级 RAG 系统
· 2 张 A800 就能跑 1 亿 token 推理。这不是实验室专属，这是创业公司买得起的成本。

说白了，以前的大模型是一个极度聪明但只有金鱼记忆的天才。MSA 想做的事情是，让它真正「记住」。

我们放 github 上了，算法的同学不容易，可以点颗星星支持一下。🌟👀🙏

https://t.co/Or3PVUv67q

[Ling @xinxialynn]

@chenyixv77 @lidangzzz 跟傻逼有什么可辩的

[Ling @xinxialynn]

@Morocaso_Sue @lidangzzz 防你妈静脉曲张，傻逼

[Ling @xinxialynn]

@OfficialLoganK 🤩🤩🤩

[Ling @xinxialynn]

Inngest claims they can streamline the backend logic but the website is too complicated.

[Poetiq @poetiq_ai]

We formed our company 173 days ago.

Today, we surpassed Gemini 3 Deep Think – at half the price – using our novel meta-system for learned test-time reasoning.

Read how we redefined the Pareto frontier: https://t.co/gytY9OA8Z0

[Alex Finn @AlexFinn]

This is it. The most important video you'll watch this year.

ClawdBot has taken X by storm. And for good reason. It's the greatest application of AI ever

Your own 24/7 AI employee

In this video I cover how it works, how to set it up, and why I think we should all be nervous:

[Ling @xinxialynn]

Do you remember when you joined X? I do! #MyXAnniversary

[Ling @xinxialynn]

When the product is strong enough, even a terrible website can’t stop people from using it. #digitalmarketing

[Ling @xinxialynn]

@fba checklist, thx~

[Ling @xinxialynn]

is deepL's business model shifting from To C to To B?