#大佬观点

H3 is the first model to reach sd-level quality and, most importantly, it’s open weights. Personally I think this is one of the biggest steps forward for both the ecosystem and the tech as a whole. From today on, we won’t be the only ones here for sure.

There are essentially two main options to remedy this: 1. Find ways to perform active inference, so that the model adapts its learned program in contact with a new data distribution at test time. Would likely lead to some meaningful progress, but it isn't the ultimate solution, more of an incremental improvement. 2. Change the training mechanism to something more robust than SGD, such as the MDL principle. This would pretty much require moving away from deep learning (curve fitting) altogether and embracing discrete program search instead (which I have advocated for many years as a way to tackle reasoning problems...)

An internal version of Astra, @OpenAI’s next major model family, solved 10 major open problems in mathematics, quantum complexity, and theoretical computer science. We believe it will be a major step for scientific reasoning. https://openai.com/index/ten-advances-in-mathematics/

Most people seriously underestimate what Grok Build can do They assume an AI coding agent is only useful for building apps or writing complex software But some of its most valuable work is handling the boring, obvious tasks that quietly consume 10 minutes here, 20 minutes there, every single day For example, I gave Grok Build a SpaceX video posted on 𝕏 and asked it to extract the clearest frames and turn them into a clean collage It pulled the video, selected the strongest images, cropped them, arranged the layout and produced the final collage shown in the post below I simply described the result I wanted, then let it handle the entire process while I worked on something else And that is only one small example You can use Grok Build to: • Organize and intelligently search files across your computer • Diagnose and fix audio, display and system-setting problems • Download, install and configure apps or tools • Extract images or audio from videos • Add music, trim, compress and convert media • Resize, crop and enhance images • Batch rename and organize hundreds of files • Clean spreadsheets, merge datasets and build dashboards • Write scripts and automate repetitive tasks • Fix broken configurations and connect tools together Even something as annoying as an audio setting not working becomes much easier Instead of hunting through menus, searching forums and trying random fixes, you can explain the problem to Grok Build It can inspect the setup, identify the likely cause, change the relevant configuration and test whether the fix worked It is basically an agentic operator living inside your computer, working within the permissions you give it I also keep a simple instruction file telling Grok how I prefer tasks to be completed, including automatically opening the finished result after a successful run so I do not forget about it Configure the workflow once, and Grok keeps working the way you prefer That is what makes Grok Build so useful It is not only building software It turns all the small computer tasks between an idea and the finished result into work you can simply describe and delegate

GPT-5.4 full 在 xhigh 下得分 51，正好与今天的 Luna max 持平。GPT-5.4 的价格是 $2.50/$15；Luna 现在只要 $0.20/$1.20。换句话说，大约四个月后，OpenAI 正以约十三分之一的 token 价格出售三月份完整旗舰级的智能。

In 2001, @JeffDean and Sanjay Ghemawat did the math and realized Google’s entire search index would fit in RAM — then shipped it in a few days, and search got fast. In 2013, another napkin calculation showed that three minutes of daily speech recognition per user would require doubling Google’s server fleet. That one became the TPU. At Startup School 2026, Google’s Chief Scientist talks with YC’s @sdianahu through the thought experiments behind both, why inference hardware is the next specialization, and where two or three people in a room can still win. 00:07 — Are AI Models Already Junior Engineers? 01:44 — AI Systems That Improve Themselves 02:40 — The Google Search Breakthrough That Changed Everything 04:38 — AI Agents Will Run for Weeks 05:58 — The Napkin Math That Led to TPUs 09:20 — How to Find Breakthrough Ideas 10:25 — The AI Engineer's New Mental Model 12:33 — Why AI Is Really an Energy Problem 16:11 — Context Engineering Is the Next Frontier 19:46 — The Skill That Made AI Better at Optimization 22:13 — Why Long-Running Agents Fail 25:21 — Where Startups Can Still Beat Google 31:19 — How to Become an AI-Native Founder 36:36 — Question Your Biggest Assumptions 42:08 — AI That Builds Better AI 50:02 — Build Something That Truly Matters