Writing
Blog
Perspectives on GenAI, community building, and the future of a more open AI ecosystem. Also on Substack.
DeepSeek Harness Is Trying to Kill Claude Code the Minecraft Way
On the evening of August 13, DeepSeek did two things.
August 21, 2026
China and U.S. open-source models meet head-on: a five-model benchmark of Kimi, Qwen, DeepSeek, Meta, and NVIDIA
Open-source models have been arriving at an almost unreasonable pace lately.
August 19, 2026
Dissecting the Kimi K3 Technical Report: Moonshot AI Starts Setting Problems for DeepSeek
On the night of July 24, Jensen Huang posted his first tweet on X.
August 13, 2026
Why ByteDance Seed Refuses to Distill Other Models
On August 5, 2026, The Information reported that Zhang Yiming, founder of ByteDance, had made clear at an all-hands meeting for Seed about a month earlier that ByteDance would not use model distillati…
August 11, 2026
MiniMax H3 vs. Seedance 2.5: Neck and Neck, Worlds Apart
On July 31, video models finally had their own collision day.
August 7, 2026
Hands-on with Qwen3.8-Max: Chinese Models Are Moving from "Can Write Code" to "Can Deliver Tasks"
If an AI asks you to come back every ten minutes to check on it, of course it is useful.
August 6, 2026
Day 1 for Qwen3.8: OpenRouter, OpenCode and More Are Already On Board
On August 3, Alibaba released Qwen3.8, its next-generation foundation model.
August 5, 2026
Inside MiniMax: The Darkest Hour Behind Its M3 Launch
At the most tense moment, the internal argument arrived right on schedule.
July 20, 2026
Recent AI Project Roundup: UXBench, MemLens, RoadmapBench...
The Next Generation of Models Needs the Next Generation of Benchmarks
July 14, 2026
GenAI Assembling Eval Eps. 1 | “AI Predicted 98% of the Exam”? We Had 8 AIs Predict the Gaokao Math Paper
Every year, in the month before China’s national college entrance exam, the Gaokao (China’s notoriously grueling, high-stakes academic exam that determines university admissions for millio…
July 7, 2026
Chinese Open Models Are Becoming the Practical Choice for AI Builders
Not every product needs the strongest model. For many real-world AI workflows, Chinese open models are starting to win on cost, access, and control.
July 1, 2026
CVPR dispatch: Alibaba Cloud cracks the agent problem
Can agents handle real-world input, run at scale, and deliver something usable?
June 22, 2026
How to Value DeepSeek: Throw Out the Model Company Playbook
The biggest mistake you can make: treating it like a model company's fundraise.
May 14, 2026
DeepSeek Can See Now. Then Quietly Deleted the Paper Explaining How.
DeepSeek Gave the Model a Finger
May 6, 2026
Audio Is the Most Underserved Interface in AI
SUN — AI-native audio learning, built around you
April 28, 2026
GPT-5.5 Isn't Selling a Model, It's Selling the Entry Point
OpenAI’s latest model is not its best at everything.
April 24, 2026
In the Agent era, what actually becomes the moat? ThinkingAI’s Silicon Valley answer
An Agent without knowledge or methodology is a human without a soul
April 20, 2026
Speedrun’s Pivot Is the Playbook: What a16z’s Accelerator Reveals About the AI Opportunity Map
Six cohorts, 240 companies, and one clear signal: the AI opportunity isn't in building — it's in replacing.
April 17, 2026
Anthropic Didn't Ban OpenClaw. It Fixed the Math
The Token Bonanza Is Over. Good.
April 9, 2026
Delve's Capital Sin — and YC's Code of Silence
Inside the Brotherhood Logic That Powers — and Polices — YC
April 6, 2026