I have first since used a couple web based AI systems for different specific tech things and they are pretty useful if you do the extra work to verify and modify what they are telling you. I’ve found that Google Gemini AI is really good a general tech questions and configuration items (with looking it over and understanding it and modifying it of course for your specific needs) and Claude Sonnet AI is really amazing at verifying that your code is correct and other more in-depth programming related tasks (for example, memory allocation and pointer handling and file descriptor tracking). I was also trying to test and see if the AI system would learn from me or others over time and add that previous historical context for each conversation into future interactions.

I’ve been using these systems lightly whenever I get stuck and they are able to find the exact code snippets I was looking for to complete the rest of my projects – prompt engineering!

Tips:

• You need to try and give as much extra debugging info that you can for it to properly figure things out or update its answer
• It’s better to ask it small specific questions and break the problem apart into pieces and then tailor a customized solution
• There are interesting differences between asking it high-level design questions versus low-level coding questions
• It’s also interesting to ask it open-minded / open-ended questions versus asking for specific solution implementations
• It may get lost easily depending on code length, complexity, and the given language it is programmed in
• Try to connect and link the history and context in between multiple sequential questions
• Sometimes it will loop through repeated answers if it gets stuck on one specific solution
• Sometimes the quality of a given answer decreases or even may change completely the longer that you interact with it
• Sometimes you will get varying, differing, or conflicting answers if you re-ask it the same question over and over again
• Sometimes it gets stuck on one specific solution that won’t work in the end but if you try to ask it how to approach the problem from a different angle it can provide steps for you to get extra info for it to either re-fine or re-suggest its solution

Note: It’s a shame that AI resources are being used on social or political issues rather than solving technology and scientific issues to help progress humanity. I’m still amazed that I was able to feed a 1000 lines of C code to Claude as a test and it was able to determine the components purposes and the network proxy applications of it all…

Google AI seems a little more friendly and personable whereas Claude is more on task and professional ha 🙂

~