Blog
Alibaba open-sourced Qwen-Image-2.1
Alibaba has open-sourced Qwen-Image-2.1, integrating text-to-image generation and image editing into a single model while also balancing quality, efficiency, and cost.
DeepSeek-V4.1-Flash feels like it was trained to crap
Continuing from the issue in the previous post, this issue isn’t a one-off—it’s the most painful experience after several days of use.
Silicon Valley AI elites change their tune again
Following up on my previous article, "Quitting to Save Humanity: Silicon Valley's Surest Can't-Lose Moral Business," today, Silicon Valley elites have successively changed their tune.
Siri AI confirmed to use custom Gemini build
Apple Intelligence isn't even available to use yet, and here comes Siri AI again. One pie has been cooking for three years—three years, and then three more. When will this ever end?
Those using CLI Agents have started to look down on those using IDEs
This phenomenon has persisted for quite a long time, and many actions and remarks have gradually emerged that completely abandon the IDE and fully embrace CLI Agent development.
The core secret to writing a good Skill: higher-order principles, not exhaustive enumeration
OpenAI recently published a post titled "Rethinking skills and prompts for GPT-6 Astra," rethinking the skills and prompts for GPT-6 Astra.
Gemini has launched a Windows client
Gemini for Mac has been available for quite a while, but progress on the Windows client has been slow; yesterday, the client was finally released.
Quitting to Save Humanity: Silicon Valley’s Surest Bet in Ethical Business
On September 8, 2026, Anthropic pretraining researcher Jacob Coxon published a long resignation essay, essentially saying that he thinks Anthropic and OpenAI are gambling with human lives, and that current AI model research could ultimately escape human control and kill all of humanity around 2030.
DeepSeek's move just left me completely dumbfounded
After multiple rounds of internal and external testing, V4.1 Flash has comprehensively surpassed V4 Pro across all metrics, including performance, cost, speed, and total time. In keeping with our responsibility to users, after V4.1 Flash officially launches and before V4.1 Pro launches, we will route all requests for V4 Pro to V4.1 Flash and bill them at V4.1 Flash's unit price. The new pricing will take effect at 12:00 Beijing Time on September 10, 2026.