ByteDance is pushing on many fronts at once. It takes every form factor to the extreme at the product level — not only in multimodal conversational assistants (consumer-facing), cloud native, FaaS and other areas, but it's already speeding down the highway on the entry-point and ecosystem side. What struck me most was Trae Solo. Sure, it's vaporware, so I'll hold off on it for now — but why would Trae, an app aimed at programmers, get so many sessions and so much airtime over two days? I think it's wildly ambitious. Trae Solo could someday become another kind of browser, or another kind of integrated office suite. I say it replaces the browser because it will become the entry point for getting all information. I say it becomes office software because it will become the tool for producing every digital output. For me, Trae can replace Doubao, replace Coze Space, replace Manus, replace Fellou, replace any agent. MCP gives it unmatched extensibility; being a VSCode environment means it inherently has the low-level ability to run all kinds of local tools; and writing and compiling code is the ultimate fallback for whatever the above can't do. It can produce code, produce documents, analyze data — it can be a work partner and an expert in any form.

I came across an introduction to the second edition of the AI Product Manager’s Handbook online, found it quite thought-provoking, and went and got myself a copy to read. The AI Product Manager’s Handbook is a practical guide to the field of AI product management, intended to help readers systematically understand and master the key methods and approaches for building, managing, and maintaining AI products in real-world work. The book was written by Irene Bratsis, and the second edition was published by Packt Publishing in November 2024. It not only covers the fundamentals of AI product management but also digs deep into technical infrastructure, model development and maintenance, deep learning, commercialization strategy, and product design, among other areas.

Wrongly flagged by AI, how do I prove “I didn't cheat with AI”? What struck us most is this: should the way we teach evolve along with technology? If AI can already do these things so well, why assign homework that AI can do so well? Where does human uniqueness come in — shouldn't homework be built around what makes humans unique?

Qwen3 is really fast — 0.2 seconds to first token, 55 TPS, more than 220% faster than the previous QwQ-32B...

Recently, the updated Trae added support for MCP, opening up a whole new way to use it — for programmers, this is another form of Manus. Then right after, I got my hands on something new that makes digital life better: a **Fellou** that puts the browser to work for me, automatically running deep research and boosting efficiency. Both are shining examples of homegrown tech.

When using QwQ-32B, the MLX version throws this error during chat: “Error rendering prompt with jinja template: Error: Parser Error: Expected closing statement token. OpenSquareBracket !== CloseStatement.”, which you need to fix by modifying the jinja template. The gguf version doesn't have this problem. After the fix it ran successfully, as shown in the image.

Trae's China version now supports DeepSeek, and it outperforms Claude3.5-Sonnet in some scenarios. QwQ-32B's performance is close to R1, making it a good fit for local deployment. If Manus goes mainstream, it will upend existing technology, and future AI will need cloud resources to support multimodal operations and agent collaboration.

It's Chinese New Year, and everyone is talking about DeepSeek. Ruirui wants to play "Who's the Spy" but can never round up enough people, so I built an AI-powered "Who's the Spy" game bot that calls an LLM to generate character descriptions and voting strategies — but I ran into problems like unstable model output and formatting errors. After tweaking the model and the prompts over and over, I finally got the basic functionality working, and the whole process was full of funny mistakes and challenges.

I'd like to recommend AnythingLLM. Combined with LM Studio, it can run local large models as a private AI assistant. I'll share my experience trying it out for document processing and studying traditional Chinese medicine. The results are only so-so for now, but the journey from zero to one is well worth it.

Crossing disciplines into TCM while struggling with sub-health: exploring how to model it rationally with the internet and AI, and comparing Doubao against ChatGPT. Since sub-health has been bother...