Well, I’m just too old to switch. My web site even boasts that my code is “guaranteed AI-free. ‘Cage-free’ is another matter.” 
3 Likes
Not sure why you feel that way. Anybody can learn how to ask questions, even administration.
2 Likes
Sorry, don’t know Asterix. I don’t understand your point.
Yea, I don’t read children’s books. So you have lost me on that. We don’t know each other, yet you are making crazy statements about me. I don’t know whether it is flattering or insulting.
There is stuff you know, stuff you don’t know and stuff you don’t know that you don’t know. For the latter category you don’t know how to ask the questions.
Of course, anyone can learn but if the foundations are missing they you will flounder at the first serious problem.
These problems will be solved. Writing highly optimized, secure, and stable code is one of the top priorities for AI companies. Because once this goal is achieved, the AI sector will develop with unprecedented acceleration.
I don’t believe in an A(G)I that will write perfect code, but I do believe in highly specialized LLMs who will write secure and stable code even without the user (human) having to point it out. Nobody wants a system that uses vibe coding to barely produce working code…
A small Claude AI generation trick
I often test development ideas by creating a prototype as a Python script with Claude AI, including prototypes with a desktop user interface.
Since porting a window user interface from Python to Xojo (API 2) can easily exceed Claude’s generation output size, I instead have Claude generate a Python script that can be used to create a Xojo window (in text format) locally. This works very well, at least for smaller windows. Of course, manual optimisation is required later on.
1 Like
I use Gemini when I forgot how to do something, and looking at the answer is enough for me to continue what I was doing (usually no copy/paste)…
So I use it as a reminder tool, sometimes better than the documentation (faster for 100% of the cases).
I do not care (more or less) if the answer is 100 % accurate if I get the idea to know how to continue to code.
how would the AI know the quality of the source code it used as input?
Or the quality of what it’s giving us.
I don’t usually give it code , I give it specifics on what to code. Often times it does not work the first time or has issues. There’s been times where I told it to fix the broken code it generated without much specifics about the issue, and it fixes it, but typically after a few passes. That makes me wonder, why did it not just generate working code in the first place.
Still, it can be a time saver.
1 Like