https://guochengqian.github.io/blog/ Essays on AI research, model scaling, agents, and markets by Gordon Guocheng Qian. en-us Sun, 05 Jul 2026 12:00:00 +0000 https://guochengqian.github.io/blog/claude-hosted-blog/ https://guochengqian.github.io/blog/claude-hosted-blog/ Sun, 05 Jul 2026 12:00:00 +0000 A clean, free blog on GitHub Pages — likes, comments, email, and sharing to X, WeChat, Rednote, and Zhihu — built entirely by talking to an agent. https://guochengqian.github.io/blog/stock-inspiration-20260701/ https://guochengqian.github.io/blog/stock-inspiration-20260701/ Wed, 01 Jul 2026 12:00:00 +0000 META's move reminded me why "it's already priced in" is often too lazy. Expectation, confirmation, and scale are different stages. https://guochengqian.github.io/blog/next-scaling-after-model-scaling/ https://guochengqian.github.io/blog/next-scaling-after-model-scaling/ Mon, 22 Jun 2026 12:00:00 +0000 Context scaling and agent scaling are the next product frontier — and both demand far more memory bandwidth. The HBM shortage is far from over. https://guochengqian.github.io/blog/research-labs-shutdown/ https://guochengqian.github.io/blog/research-labs-shutdown/ Sat, 20 Jun 2026 12:00:00 +0000 Research labs at the giants are shutting down, merging, and shrinking. This is the organizational reconstruction that follows AI industrialization. https://guochengqian.github.io/blog/gemini-omni-architecture/ https://guochengqian.github.io/blog/gemini-omni-architecture/ Thu, 18 Jun 2026 12:00:00 +0000 Three plausible architectures for Google's latest video model, ranked — and why a Transfusion-style unified transformer is my top guess. https://guochengqian.github.io/blog/opd/ https://guochengqian.github.io/blog/opd/ Mon, 15 Jun 2026 12:00:00 +0000 OPD is not just a loss function. It is how a thousand-person lab works in parallel on the same intelligence. https://guochengqian.github.io/blog/post-training-market/ https://guochengqian.github.io/blog/post-training-market/ Wed, 10 Jun 2026 12:00:00 +0000 Two years ago pretraining used ~10x more compute than post-training. Six months ago the ratio was closer to 1:1. Now it may flip. https://guochengqian.github.io/blog/visual-generation-chatbot-accessory/ https://guochengqian.github.io/blog/visual-generation-chatbot-accessory/ Wed, 03 Jun 2026 12:00:00 +0000 Visual generation is moving from standalone apps into chatbots, agents, and multimodal operating systems. https://guochengqian.github.io/blog/agent-for-finance/ https://guochengqian.github.io/blog/agent-for-finance/ Sat, 30 May 2026 12:00:00 +0000 When an agent has million-token context and tracks policy, news, and market data faster than any retail investor, it becomes a real research edge. https://guochengqian.github.io/blog/elorian-ai/ https://guochengqian.github.io/blog/elorian-ai/ Sun, 24 May 2026 12:00:00 +0000 Notes from Andrew Dai's interview: 14 years inside Google's AI machine, why he left after Gemini, and Elorian's bet on visual reasoning. https://guochengqian.github.io/blog/hermes-agent/ https://guochengqian.github.io/blog/hermes-agent/ Mon, 18 May 2026 12:00:00 +0000 My setup for an 8am daily briefing, 10pm summaries, paper writing, blogging, and remote work from my phone.