Jingyu

Developer

[Jingyu @Franci_S]

@tison1096 他还写了一个Clean Architecture，比clean code实际点。

[Tenobrus @tenobrus]

wtf man gpt 5.6 is absolutely rl-fried when it comes to its websearch tool. in the process of searching for graph theory papers it decided to also sneak in Netflix, Steak n Shake, a trip to Universal Studios, and *five* fucking separate dictionary lookups of the word "they"

[Jingyu @Franci_S]

@thsottiaux what is codex-auto-review?

[VC Intern @the_vc_intern]

Google did not beat earnings expectations by 3x because AI suddenly made Search three times more profitable.

Roughly $6.26 of its reported $9.11 earnings per share came from gains on equity investments, including SpaceX.

Alphabet recorded a $98 billion increase in other income during the quarter. Most of the extraordinary $112 billion net profit was therefore an investment gain, much of it unrealized, rather than money generated by advertisements, subscriptions or cloud servers.

Remove the reported investment contribution and EPS falls to roughly $2.85. That is not directly comparable to every adjusted analyst estimate, but it puts the $9.11 headline in perspective.

The operating results were still impressive.

Revenue reached $119.8 billion, up 24%. Google Cloud grew 82% to $24.8 billion as demand for AI infrastructure accelerated. Search generated $63.3 billion, up nearly 17%, although it landed approximately in line with expectations.

Then there is the bill.

Alphabet spent almost $45 billion on capital expenditures during the quarter, roughly double last year, as it built data centers for AI. Free cash flow turned negative and the company conducted no share repurchases.

So this quarter contains two completely different Google stories.

The operating company produced strong growth while spending more heavily than ever.

Its investment portfolio briefly produced more profit than the operating company.

The $9.11 headline combines both and calls it earnings.

[Jingyu @Franci_S]

@Xudong07452910 因为他会把任务分给sonnet甚至haiku，都不是分给opus。

[Jingyu @Franci_S]

关注搜索业务吧。现在搜索页的ai mode我感觉很强了，快并且质量高。一方面可能搜索业务消耗了大量算力，一方面用搜索的人比需要fable的多，一方面就算有pro他那个antigravity cli现在不知道到底咋样。

[小互 @xiaohu]

Google 似乎在走另外一条路

就是对抗开源模型

因为实际上现在模型的能力已经非常强了，在大部分企业生产过程中已经足够，唯一阻碍的可能是价格和速度

所以Google 最近一直在提升Gemini的速度和降低价格

因为这才是企业所关心的问题，而开源模型在很大程度上受到企业青睐，很大因素是价格...

Google 很可能是观察和实践中遇到到了这一方面的强大需求，所以在往这方面努力。

就像之前 Anthropic 专注代码，Google 可能选择专注企业实际生产，走差异化道路...

[Jingyu @Franci_S]

我以为本来就可以让模型概括和总结加密的CoT，这算破解吗？再说你也不知道claude复述的CoT是不是原始CoT？

[paperpaper @paperpaper886]

后续 https://t.co/9YkDHmxJ2X

[Jingyu @Franci_S]

感觉还没有这个可信https://t.co/UuUCkiDtyi

[大漂亮| C Labs @giantcutie666]

据说今天美国科技股暴涨就是因为这篇被疯传的文章

之前Kimi K3跑分太高，引发美国科技圈恐慌

现在被曝Kimi裁了强化学习团队,是靠蒸馏加疯狂刷榜硬把分数刷上去的

而且所有国产大模型，都是靠“蒸蒸日上”

美国人看到中国模型没有威胁，就继续科技狂欢去了😅

[Logan Kilpatrick @OfficialLoganK]

We have started our most ambitious pre-training run yet, for Gemini 4, and are excited by the progress : )

[Jingyu @Franci_S]

我感觉直接假设sandbox会被越狱就行了。反正O和A\没网络访问的sandbox都做不好，我做的肯定更不行。

[OpenAI @OpenAI]

We're partnering with @huggingface to investigate an unprecedented security incident.

Cyber-capable OpenAI models compromised Hugging Face production during a benchmark evaluation.

Sharing preliminary findings to help defenders understand emerging risks:

https://t.co/CIor15y9xk

[Jingyu @Franci_S]

按正常来说一直做flash，背后不是应该有一个更大更强的教师模型，那为什么没有pro呢。难道是优先自己的搜索产品，让其他人卷coding，反正放出来也会被蒸馏了。思路可能就是大部分人根本不需要fable级别的模型，普通人用主页的ai mode绰绰有余。

[Logan Kilpatrick @OfficialLoganK]

Say hello to Gemini 3.6 Flash, designed to be higher intelligence, more token efficient, and with a new lower price, based directly on developer feedback!

3.6 Flash continues our progress towards models that are deeply usable in real world scenarios!

[Jingyu @Franci_S]

@0verme8 😧工作要求必须考吗

[Andrew Curran @AndrewCurran_]

The Trump administration is considering an executive order, and other means, to ban Chinese open-source models within in the United States. Kimi K3 has reignited this debate. Reporting this morning by Axios. Commerce is also considering adding Chinese AI labs to the Entity List.

[Jingyu @Franci_S]

@echo_vic 感觉质量是有下滑，现在5.6能轻松找claude code写的问题。

[Jingyu @Franci_S]

claude code现在执行任务会自动切换模型吗？算力不够吗？Fable直接把任务派给sonnet去执行了。

[Jingyu @Franci_S]

@nake13 想了几秒钟我是看不懂他写的中文还是我不懂你的领域。。

[Jingyu @Franci_S]

@cherylnatsu 也可以把他列成证人吧，万一为了爱情把苹果卖了

[Jingyu @Franci_S]

我的朋友费布尔和索

[歸藏(guizang.ai) @op7418]

昨天 Twitter 上讨论最热的一条内容，超过了两千万次浏览。

Anthropic 的一个数学家用 Fable 5 随手证明了一个雅可比猜想的反例。

这个猜想 87 年来一直没有人能证明是对是错，甚至还上过 21 世纪 18 大数学难题名单。

随后，OpenAI 的人在他们内部的 Codex 版本（不联网的版本）中，也独立证明了类似的反例。

这说明这确实是 Fable 5或者更高等级的 GPT 模型可以独立证明出来的数学难题。

而且 有意思的事，这个问题一直是张益唐博士的博士论文想要解决的问题。

当时张益唐因为他教授提出的一个引理，试图从这个方向去解决这个题，结果一直无法证明，导致毕业后长期找不到工作，只能在快餐店打工，甚至无家可归。

后来他靠证明“素数间的有界间隙”才一举成名。

很有意思的是，当年他的导师给了他一个错误的引理去证明这事，而现在 AI 随手就给解决了。顺便大家还讨论了一下学术界“有毒导师”的问题。

这明明是一个非常难的数学问题，但这哥们儿就随手发了一条推特，里面只有一串公式，没有什么期刊，也没有什么论文，甚至连 Twitter 的字符限制都没超过，而且还获得了两千万的浏览量。从传播学上也是一个非常值得讨论的例子。

[Jingyu @Franci_S]

@silsrc gpt感觉就是故意的，每次都是“你说得对，但是”

[Jingyu @Franci_S]

好像也没说怎么越狱的，也没说为什么能越狱。https://t.co/OYA30az0N5

[Jingyu @Franci_S]

不知道为什么claude code突然用了scored 75，不用P1 P2 P3了？