The AI e-commerce war between Ali and Byte was really lively. Qianwen and Taobao get through are essentially an upgraded version of AI shopping guide. There are no new tricks in technology, but just packaging the recommendation algorithm into an interactive dialogue. But in business, this move is tough enough to directly block Byte's AI e-commerce ambitions. The question is, can this kind of conversational shopping really be more efficient than traditional search and recommendations? I suspect that most users still end up directly comparing prices and reading reviews. The AI assistant is at best a fancy entry point. The most practical use may be to help merchants generate personalized marketing words, which can increase conversion rates.
The direction of medical AI has finally begun to be pragmatic. In the past few years, everyone has been talking about drug discovery, but now smart people are starting to make document processing and project management tools. This is the real pain point-paperwork in the pharmaceutical industry can drive people crazy, and it is normal for clinical trial plans to be revised to version 30. But the biggest challenge with this tool is compliance. Medical data cannot be fed to AI casually. If you do it well, you can indeed become a cash cow business. After all, the most important thing pharmaceutical companies lack is budgets.
In terms of hardware, NIO's power exchange station layout speed is indeed astonishing, but whether this business model can continue remains a question mark. Every station is a heavy asset and its maintenance costs are so high that it is expensive. It is no problem to rely on capital to support it now, but it is hard to say whether it will be profitable after the subsidies recede. As for the manned mech, it is purely a PR gimmick and cannot be put into practice at this stage. The investor is probably just playing with the founder.
The most interesting thing is the changing trend in the developer community. Now there are a bunch of skills libraries on GitHub trending for teaching AI to write code, which shows that the pain points of large models in engineering practice have been clearly exposed-it writes code like a freshman and needs to be taught by someone. These skill libraries are essentially making up lessons for AI, which is quite ironic. AI should have taught us to write code, but the result was reversed. However, this kind of crowdsourcing training is indeed better than working behind closed doors. At least the problems are all on the surface.