最近几天，中国低成本大语言模型深度求索（DeepSeek）欧美AI圈引起了不小的震动。据悉，来自杭州的初创企业深度求索1月20日发布DeepSeek-R1，该模型在测试表现、训练成本和开源开放程度等多个基准测试中均超越“ChatGPT之父”美国OpenAI公司的最新模型o1，但成本仅为o1的三十分之一。
The pipeline was very similar to icon-to-image above: ask Opus 4.5 to fulfill a long list of constraints with the addition of Python bindings. But there’s another thing that I wanted to test that would be extremely useful if it worked: WebAssembly (WASM) output with wasm-bindgen. Rust code compiled to WASM allows it to be run in any modern web browser with the speed benefits intact: no dependencies needed, and therefore should be future-proof. However, there’s a problem: I would have to design an interface and I am not a front end person, and I say without hyperbole that for me, designing even a simple HTML/CSS/JS front end for a project is more stressful than training an AI. However, Opus 4.5 is able to take general guidelines and get it into something workable: I first told it to use Pico CSS and vanilla JavaScript and that was enough, but then I had an idea to tell it to use shadcn/ui — a minimalistic design framework normally reserved for Web Components — along with screenshots from that website as examples. That also worked.
，这一点在WPS下载最新地址中也有详细论述
* - this.stack：单调递减栈，存储「价格对应的索引」，栈顶→栈底索引对应的价格单调递减
"updatedActiveQuests": [