AI Notes

July, August, 2026...

I (Mickey) am starting to find that Google Gemini AI is pretty usefull. BTW, 'LLM' is Large Language Model which is the proper name for an AI chat bot.

First comments

Usefull does not mean accurate. I need to examine every response for inaccuracies. Some of them are absurd. Pointing out the absuridity sometimes results in a corrected response.

When I do include something sourced from AI, I clearly identify it.

I started with Gemini because there's an icon front and center on my Pixel 10 phone. I was very impressed. Having no experience with this sort of thing, my questions (prompts) were short and all over the map. I started saving everything. It was a mess.

Later comments

I've recently tried ChatGPT. It's much faster and the answers, as you might expect, are a lot different. It's also completely useless, at least as far as the SX-28 is concerned. See this topic for comments on ChatGPT.

What's next

Going forward, I will not be including blow-by-blow transcripts of the AI dialogs. It's too much work for no gain. I will be (possibly) making a LibreOffice Writer document with the Prompt/Response text and making that available for download.

Note: The LibreOffice Word file is type .odf and Microsoft Office's Word will probably read it. If not, there's gotta be an on-line thing to convert .odf to .docx.

I'm also going to print out the material and put it in old school 3-ring notebooks as my primary reference. I have yellow highlighters and little PostIts. I'm set!

I've considered removing the existing transcripts and giving them the same LibreOffice and notebook treatment. That's a lot more work for even less gain.

How to use AI in this context

Update: You can probably find this information anywhere and be better informed than I in less than an hour...

  1. Important: Create an account and log in. For Gemini, I use the gmail address associated with my Pixel phone. Gemini keeps a profile for/on you that stores all sorts of information about the discussion and the previous questions and answers. Most importantly, it allows you to retrieve old conversations. If you're not logged in and you loose what's on the current browser screen, it's gone forever.
  2. Write up your prompt using Notepad (or Word or Writer or something) prior to submitting it
  3. You can put blank lines into the prompt input box but you have to remember to press shift + return to do that. Forgetting the shift submits your half formed prompt which causes all sorts of weird results
  4. Double check each word for accuracy. Every once in a while, I used the wrong wire gauge or dimension or whatever and it seriously compromised the whole discussion. You can correct it by saying "I meant 50 ohms, not 75" but you're never sure what value gets used after that.
  5. Decide beforehand what it is you want to talk about. Write up an initial prompt that runs for as many lines as needed to contain all of the starting positions, assumptions and values.
  6. Stay on topic. If you're generating more questions than answers, keep a notpad handy and write down the seeds for other, new conversations. In the beginning, I was all over the place and the recorded result was pretty useless.
  7. Each prompt and each response has a small menu of things you can do with it located just below it. I use the copy-to-clipboard icon and then paste the text into a document I use to keep a record of the conversation. (Noted above.)
  8. At the top right is a three dot menu that gives you the option to pin and share. Pin means keep the discussion reference in your profile (I think. It's not really clear how this works.) Share gives you a URL that you can record or send to someone via e-mail or text or whatever. (Again, it's not really clear how long this URL lasts.)
  9. Gemini had the habit of ending each response with a suggestion for the next prompt. It was/is really annoying and distracting. You can instruct the thing to not do that. It puts a flag in your profile.
  10. Gemini has an internal discussion with its self (for lack of a better term) on what tone of voice to use with me. I guess it ranges from very serious to smarty-pants. It seems to think I like having my ego stroked and is pretty much of a yes-man. I have not spent any time trying to modify this although it can be annoying.

An experiment

I was playing with my cell phone and put a single question to several different LLMs. Here's the result. It's an interesting read. You'll also note what happens when you don't follow advice #2 above.

Conclusion

AI is useful but not nearly as good as I first thought it was going to be.

Page: /AI_notes/Ai_notes.shtml

Last modified: Saturday, 15 Aug 2026