I've been finding it helpful to prompt not to emit any code except for snippets to illustrate a point. Then I hold a discussion of a proposed approach. Only then do I ask for some code. I think using the chatbots can provide some of the benefits of pair programming. But you need to be involved closely for the best results.
Remember the old saying, that you need to be twice as smart to debug code as to write it. So if your code is as clever as you can make it, then you are not smart enough to debug it. On Friday, June 27, 2025 at 11:11:19 PM UTC-4 mys...@gmail.com wrote: > On Wed, Jun 11, 2025 at 6:29 AM Thomas Passin <tbp1...@gmail.com> wrote: > >> I just asked ChatGPT an involved question about its capabilties for >> developing large and complex software. > > ... >> ChatGPT>> >> ======== >> > "The short answer is: *no, not independently, and not reliably*—at least >> not in the way such a system would be developed by a professional >> engineering team operating under disciplined systems and software >> engineering protocols. >> > >> - >> >> I lack real-time testing feedback, so UI behavior is hypothetical >> unless you test and report back. >> >> >> - >> >> Accessibility, internationalization, responsiveness, and >> platform-specific edge cases are typically overlooked unless explicitly >> specified. >> >> >> >> I can assist in scripting or describing tools to manage these tasks, but *I >> do not act as a team or development environment*. You would need human >> oversight to manage everything from software configuration management to >> risk assessment. >> ------------------------------ >> >> I am a useful *component* in the development pipeline for a large-scale >> system, especially for: >> > >> - >> >> Boilerplate code generation. >> - >> >> Requirement interpretation (in chunks). >> - >> >> Test scaffolding. >> - >> >> Code review. >> - >> >> UI prototyping. >> >> But I cannot independently develop, verify, or maintain a software system >> at the scale or complexity of AWIPS or any comparable mission-critical >> platform. Any claim to the contrary is speculative or based on trivial >> examples, not grounded in the realities of enterprise-scale software >> engineering." >> > >> End ChatGPT> > > So far in my tests, Gemini 2.5 Pro seems to be giving me interestingly > useful BASH results. Here's a deep dive into the coding of a moderately > useful cpu p-state control script. > https//github.com/mystica555/cccpu > > Is the communist theme a bit too much? would you suggest something else? > I'm open to suggestions, and have a lot (a _LOT_) of images that came out > of 2.5 pro chat mode. So much I had to slow down and can't use 2.5 pro > again until tomorrow. > > I could link a gallery, if anyone is interested. > > I think i figured out a decent way of importing the files into git the > first time as well, only about 4 iterations w/gemini on a quick script for > that and then fixing it to copy ccccp.x.y.sh to ccccp.sh so the version > reflects the same file.. before this it naively imported the files with the > versions still as part of their filenames. > > For now, I figure if I can be a design director and critiquer, fixing > randomness that pops up but mostly just kindly asking tasks of it, gemini > does a good job. Trying to find the right balance between prompt length and > desired outcome has been an interesting learning curve, that I still think > I am gong up. > > Mike > > -- You received this message because you are subscribed to the Google Groups "leo-editor" group. To unsubscribe from this group and stop receiving emails from it, send an email to leo-editor+unsubscr...@googlegroups.com. To view this discussion visit https://groups.google.com/d/msgid/leo-editor/07cd1eb3-a0ca-4e80-875c-3846df709e9fn%40googlegroups.com.
