10-30-2025, 11:20 AM
With something like QB64, unless you are using an AI trained specifically for QB64, you will almost always have hallucinations. The standard Gemini doesn't know enough about QB64 code or syntax to provide anything useful. It knows that QB64 can use more modern stuff, it knows you can use external libraries, and it knows that more modern keywords are given an underscore. So, it decides to make up stuff. That's why I like using NotebookLM for this kind of thing. Its scope is way smaller than Gemini. It only really knows about what you give it and will flat out give up trying to respond if it can't find anything in the sources to help. Gemini is best used for things that are common or popular, so it has a higher chance of knowing how to respond. NotebookLM is best used for things that are more niche. I once tried making a Gemini Gem (similar to a custom GPT) but it only allows 10 sources and it does not strictly adhere to any rules, just like ChatGPT's custom GPTs.
The noticing will continue

