While DeepSeek is interesting, when it tells me 'Too busy; try again later.", that makes it easier to go back to ChatGPT, where I very rarely have to try again. Of course, if I’m using the DeepSeek-R1 locally with Ollama, then I don’t have the busy issue. But, I am working with a much smaller LLM in that case.
Speaking of AI in general, I do find it interesting that, if you can be extremely specific in what is required, AI can do a somewhat decent job of creating a starter project for you. Of course, if you’re not sure you are specific enough, you can have it review your instructions and suggest options for clarification. But, you still need to have thought through a decent starting set of features. Of course, even then, there will still be cases where you have to correct the AI hallucinations. For example, when fixing resulting errors, it often suggests a fix that, while correcting one thing, will remove other required functionality. Or, it may get in a cyclic loop of delete A and add B only to later have deleted B and added A back, resulting in the original error all over again. This is partly related to limited context lengths. It cannot always hold sufficient scope simultaneously to make correct suggestions. So, you still need to keep on top of it and know enough to write code the old fashioned way, by using your own mind and with your own fingers on the keyboard.
(DeepSeek and 
I’m curious what will happen. 
