D

I love AI, AI loves me.

[D @quantduck]

世界可以分成好人坏人阵营吗?

[D @quantduck]

sol 6.1 发布说明我猜对了。
而且我最近也在想，如果我必须要用最顶级的模型开发，说明我的项目有问题。就像一个公司你需要100个天才这是不合理的。

[D @quantduck]

近期AI基本到顶了，未来就是降价。 AGI也不会诞生在anthropic和openai,因为船大难调头，即使知道AGI路线也不会选择。

[D @quantduck]

近期AI基本到顶了，未来就是降价。 AGI也不会诞生在anthropic和openai,因为船大难调头，即使知道AGI路线也不会选择。

[D @quantduck]

一个项目第一天和第一百天目标完全不一样，根本没办法提前设计。看人类进化就明白，好东西是进化而不是设计。

[毒猫猫 | 量化摸鱼师 @NekoStranding]

好业务一般都是屎山开局

如果一个事真赚钱，那肯定是一堆人火急火燎的往上扑。不会只有懂架构的人在做，也不会给团队好好修理架构的时间

[D @quantduck]

还需要从游戏规则去分析https://t.co/z1sKcINSMU

[NIK @ns123abc]

🚨BREAKING: Jensen Huang just EXPOSED Dario and Altman’s "Rogue AI" grift to avoid getting sued into oblivion under EXISTING law

“Don't let this doomsday narrative cause somebody to relieve them of the laws that currently exist.

Go and read between the lines. They're actually not asking for more laws; they're asking to be relieved of the laws we do have.”

ABSOLUTE TRUTH NUKE

[D @quantduck]

当你看不懂一件事的时候，就用最基础的经济模型去判断。他面临什么，他在做什么，不是听他说什么

[Dario Amodei @DarioAmodei]

We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so.

Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our systems, so that they can verify adherence to our safety measures, report on incidents, and assess models’ alignment during training.

You can read the full post here: https://t.co/OGyPb7yaYt

[D @quantduck]

不要用���德的滤镜去评判一个资本家

[D @quantduck]

感觉AI已经到了平台期了,其实现在的大模型和GPT4真的有很大的智力提升吗?还是有幻觉,还是会遗忘,还是不能持续学习.
我已经很久不使用最顶级的模型了,各种模型进行组合效果更好.

[D @quantduck]

世界的高层从古至今都很相似,他们没有太多情绪和分歧.即使战争再激烈也能亲切的握手.

[Floro S. @sflorimm]

remember this:

competition happens at the bottom.

the people at the top are collaborating.

[Simon Martinelli @simas_ch]

Spec-driven development is not one thing. There are three levels.

Birgitta Böckeler from Thoughtworks describes them in her article:

Spec-first: write a spec, generate the code, throw the spec away.
Spec-anchored: the spec stays in the repository and evolves with the code.
Spec-as-source: humans only edit the spec, the code is regenerated.

Where are the tools?

Kiro and spec-kit are spec-first.
BMad is spec-first with good planning documents on top.
OpenSpec is spec-anchored, with delta specs that merge into the main specs.

And the AI Unified Process? Also spec-anchored.

Requirements, use case specifications, and the entity model live next to the code for the whole life of the system. Tests are linked to use cases with @UseCase. Change the spec first, then the agent changes the code.

I rarely type code myself anymore. So is this level 3? No, and on purpose. The code is not regenerated, the agent evolves it. That is the only way brownfield works. Nobody regenerates a 15-year-old ERP from a spec.

You get most of the productivity of level 3 without depending on a non-deterministic compiler. And you keep a team that understands its own system.

Full article: https://t.co/W8y1zxThXt

#SpecDrivenDevelopment #AIUnifiedProcess #RequirementsEngineering #SoftwareArchitecture

[Theo @droid_35719]

https://t.co/rSoZNsALNJ

[D @quantduck]

等国产400块

[clem 🤗 @ClementDelangue]

https://t.co/KXKqDSkNJj

[Factory @FactoryAI]

We built the world's most advanced program reverse-engineering system. It's model-independent and demonstrated improved performance for every frontier model we tested on long-horizon software tasks by giving agents an independent definition of done.

New Factory research:

[D @quantduck]

效率增长率<由效率带来的服务需求增长率⇒总资源消耗增加

[Paul Bonnet @PaulBonnet]

In 1852 aluminium cost twice as much as gold.
Then the price fell over 99.9%. Yet it built one of the largest materials markets on earth.
What it says about AI tokens: https://t.co/5Y04wrz8mj

[D @quantduck]

😆

[Zack Korman @ZackKorman]

I support data centers unless they're built or used by Anthropic, in which case they take too much water and give you cancer

[D @quantduck]

确保我们掌握技术，而不是被技术所掌控

[Pi @pidotdev]

This blog post was written for those who may be curious to know what an agent harness is, but don’t, and might not know where to start. Read the full post here

https://t.co/pUnUnQvEHJ

[徹言 @iamcheyan]

最近比较闲，一边学日语，一边重新研究以前感兴趣的开源项目。
但现在已经很难静下心从头读代码了，毕竟 AI 读得比我快。
于是换了个方法：让 AI 读一个历史悠久项目的 GitHub 提交记录，再结合 release、issue、论坛讨论、RSS、新闻和当时的时间背景，整理成一部项目编年史。让它像写小说一样地，讲这个项目如何从 0 到 1 被实现出来的，没事就打开看看，老有意思了。

https://t.co/0b8bVDAWTN

[Ben Dickson @bendee983]

Running multiple AI agents in parallel creates a communication bottleneck, where models can't share their findings mid-step and need to wait at specific steps to coordinate and communicate.

AgentRadio solves this communication bottleneck with a lightweight but efficient framework that is compatible with existing agentic harnesses such as Claude Code and Codex CLI. At the heart of AgentRadio are three primitives:

- create_thread (create a new message thread between agents)
- send_message (send message to thread)
- wait_for_mention (wait for message on thread)

The only requirement is for the target harness to support background processes for the wait_for_mention primitive to enable “passive awareness,” where agents can continue their primary tasks while passing messages and updating their knowledge in the background.

In practice, this means that, for example, if an agent makes an important discovery in the middle of one of the steps, it can communicate the finding to other agents in the middle of the current steps and the other agents can update their planning accordingly, without waiting for specific milestones.

In experiments, with equal compute budgets, AgentRadio beats single agents as well as multi-agent systems without passive awareness on complex tasks such as codebase understanding (SWE-Atlas QnA) where agents need to do exploration by not only reading the code but also running tests and using tools.

[D @quantduck]

要提升代码管理能力，不在于逐行细读代码，而如同管理团队，重在把握整体而非盯死每个个体。

[𝗠𝗶𝗰𝗵𝗮𝗲𝗹 𝗞𝗼𝘃𝗲 @michael_kove]

Reading the code is purely ego thing.

If you have strict guidelines and enforce code architecture - SOTA models will produce code that adheres to those rules (ie "Best practices")

AI will write tests, most of you will never.

AI can review itself, find repetitions and issues.

In rare cases AI would over-engineer solution, that's why planning mode is so important.

It's also concert of

"there's a SKILL for that"

I don't oneshot.

I iterate. Plan. Review. Document. Build. Test. Review. Test. Audit.

All are skills, workflows, mostly automated.

Reading the code becomes informative practice. Because we're at the point of 95+% deterministic outputs.

Sufficient for pretty much all of your projects.

A sliver subset of development work with unpopular languages or problems models never seen in training...

But it ain't you, bruh. Most of devs on here are not working with that. It's truly rare.

[D @quantduck]

限制 ai 的边界和上下文

[Uncle Bob Martin @unclebobmartin]

When using agents, ensure that everything that can be deterministic, is done with a deterministic tool. Don't try to get the poor agents to follow a deterministic process.

[D @quantduck]

我相信他们当中有人能活到200岁✅

[D @quantduck]

如果以200岁预期寿命,团队很年轻.

[Jeff Dean @JeffDean]

Announcing Discovery Loop!

I am very excited to announce that, along with my longtime friends and collaborators @Sanjay_Ghemawat, @OriolVinyalsML and @quocleix, we are founding Discovery Loop (@DiscoLoopAI), a Public Benefit Corporation whose mission is to automate machine learning, science, and engineering to accelerate discoveries and progress. The four of us have worked together for 14 to 30 years, and have helped build some of the world’s most used products, infrastructure and AI models, and we’re excited to turn our attention to this ambitious endeavor.

♾

Learn more at: https://t.co/Rv3LMdLluK