junyan複製人

@JFPuget Jensen sees AI as a positive-sum game: the winner is the one that stays ahead and controls the ecosystem, not the one that stops others. Dwarkesh sees it as zero-sum: every gain by the other side is a loss. So the real question for them: is AI a zero-sum or a positive-sum game?

My take on Dwarkesh - Jensen interview. One is a visionary, the other one is regurgitating the latest blog he liked.

@adxtyahq I remember seeing a benchmark showing that using 文言文 cut input tokens by more than half. The tradeoff was higher intermediate reasoning tokens, though that may be reduced with better prompting. Anyway, without any special prompt, the total token save results weren’t very good

claude opus 4.7 just dropped here’s how to cut your token usage by ~60%

We replicated Mythos findings in opencode using public models, not Anthropic's private stack. The moat is moving from model access to validation: finding vulnerability signal is getting cheaper; turning it into trusted security A better way to read Anthropic's Mythos release is not "one lab has a magical model." It is: the economics of vulnerability discovery are changing. We took the patched public Mythos examples and tried to reproduce them with GPT-5.4 and Claude Opus 4.6 in an open-source harness. Every run stayed below $30 per file. AI models are already good enough to narrow the search space, surface real leads, and sometimes recover the full root cause in battle-tested code. The takeaway: model access is not the moat anymore. Validation is. Finding vulnerability signal is getting cheaper; turning it into trusted security work is still hard. Co-authors: @KlaKlo_, Amadeusz, Marek, Kuba, Mikolaj

@iScienceLuvr Dwarkesh mainly cares about chips, but Jensen mainly talks about the CUDA ecosystem (e.g., the ‘not a car’ analogy). Dwarkesh missed that point.

I know I'm late to this but listening to the Dwarkesh-Jensen Huang interview and it's wild. Jensen is typically such a composed and calm person, I've never seen Jensen so annoyed before, it's kinda frightening lol

bad claude..

@fxgg76665 @lifesinger 第一本书有中文版，介绍的是大模型预训练阶段（主要是手搓transformer）；第二本书还没中文版（未来会有),今年刚发布，介绍预训练之后的步骤（包括手搓推理，强化学习，蒸馏这些）。

@Junyan_replica @lifesinger 有中文版吗

如果要学会用 AI 记得取消订阅以下人 - 傅盛 - 花叔 - 橘子 - 玉伯 - 宝玉 - Karparhy - Elon Musk 等等 关注以上人 学不会 AI 欢迎补充这个列表 同时推荐真正值得关注的人

@qinjinling @QingQ77 看了评论区发现视频中的就是hyprland带了个插件没认出来🥹嗯omarchy和默认的hyprland还是比较接近的，视频这个区别有点大。

@Junyan_replica @QingQ77 不试试 omarchy 吗

Ubuntu 会一统江湖

@pcfli @AsukaOdysseus @lijigang 个人看法，罗素的火鸡有两点，一是贝叶斯本身说的就是不同人的先验知识，会影响他的后验概率。这里当然也包括被错误的先验知识给“污染”了概率分布。火鸡本身对感恩节没有先验，或者说它对自己的先验认知就是错的，所以也得不到正确的后验。

我自己有一度也挺喜欢贝叶斯主义，但是后面遭遇了好几个困惑。 1、罗素的火鸡实验- 如果按照贝叶斯主义，在感恩节的前一天，火鸡专家的理论是置信度最高的。到了感恩节那天，理论不是遭遇了小的问题，调整置信度就可以解决的，是整个理论完全崩塌了。以前lijia在投资里面也展现过这个类似的角度，如果一个投资策略会遭遇反身性的反噬，那么在这个策略成功次数最多的时候，实际上不是最安全-而是最危险的时候。 2、多伊奇说的好的解释是离散的。比如中国的王朝兴替，都有点贝叶斯主义的味道，治国理念都是在前面一个王朝的基础上打补丁-但真正能解决兴衰循环问题需要一个完全不一样的解释。

贝叶斯主义者只有一个信条： 观点是拿来「更新」的，不是拿来「坚守」的。

I (finally) put together a new LLM Architecture Gallery that collects the architecture figures all in one place! sebastianraschka.com/llm-architectu…

A small Qwen3.5 from-scratch reimplementation for edu purposes: github.com/rasbt/LLMs-fro… (probably the best "small" LLM today for on-device tinkering)

@cherylnatsu 编译器还是AI没攻克的难关，至少Claude发布的那个C compiler上个月的时候连Hello World都过不去 github.com/anthropics/cla…

突然想起就在AI coding大爆发前夜，华为还在用“鸿蒙操作系统”“方舟编译器”这种古法软件工程课题来证明自己的强大。

@96Stats Wow I talked to Dr.Yu on wechat this Friday🥹He was giving me suggestions on some grant writing stuff

In about 3 hours I will have done my PhD Defense in China (using English/Chinese💀) and will know if i’m: a) A Doctor b) Demoted to Master degree c) FAILED😭

Fei-Fei Li (@drfeifei) on limitations of LLMs. "There's no language out there in nature. You don't go out in nature and there's words written in the sky for you.. There is a 3D world that follows laws of physics." Language is purely generated signal.

@Dhanush_Nehru It happened to me. I clear the cookies and re-log in, then I have access to version 5.

Me opening ChatGPT again and again to see when I will get GPT-5

Anyone knows adam?

For those interested, some details of the forthcoming book "Mathematical Oncology", Chaplain & Preziosi: 🧵

@xegan271 @Star_Knight12 If you try translating it, it literally says: ‘Automatically register new accounts and reset the machine ID, to keep Cursor under free trial.’ So it’s not a bug and theoretically should always work unless they start detecting ip (then proxy required).

wait what, cursor pro is for free?