On Wed, May 13, 2026, 20:11 Thomas Passin <[email protected]> wrote:
> So far I have had a 100% failure rate in trying to get Claude or ChatGPT > help solve small but tricky problems. They have burned up a large number of > free tokens and made many suggestions but haven't solved the problems. > > They also don't seem to have a large enough context window to be able to > maintain the overall picture of large threads. So I don't see how they > could be able to create a decent large system design. Just complete enough > specifications for one component would probably overflow the context window. > > Given that we know that good quality and reliable operation cannot be > achieved by testing alone, I don't see how widespread adoption is going to > be anything but a long term software nightmare. All these systems that may > *seem* to work but watch out when they hit edge cases... One thing you might want to try is a proper coding harness for the chatbot. I find that using Google's Gemini CLI calls multiple sessions for individual subtasks which allows for the context window of each individual subtask to be dedicated for it alone and you don't end up with a massive chat history that the LLM can't keep up with 5 or 6 messages into it because it's done so much work and it keeps seeing the same session and the same history every single time you send a new message. By splitting the coding task up into many small subtasks this works a lot better then just talking into the website. It chooses which model to run based on the task complexity and comes up with a cohesive result in the end I find, more often than not. Just my 2 cents. 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 [email protected]. To view this discussion visit https://groups.google.com/d/msgid/leo-editor/CAN%2B%2B4hGdNJYbB7bRSkhdYcRTqwvidtA1Ayt1NR9PJpr%3DFMyNkQ%40mail.gmail.com.
