Quote:
Originally Posted by Hrycaj
I have never used Claude so I do not know how it differs than other AI's like Chat and Gemini. Through what perspective does Claude look at the information? I have done the whole assistant Gm thing with other AI's but you have to tell it what perspective to look at the information through. Basically I have to tell it which stats I value from stats to personalities before I feed it information.
Example: If I have Gemini evaluating pitchers in a game from the 50's with a 2026 mindset they will hate all of them because the velocity in the 50's was nowhere near what it was today.
Which brings me to my next thought. When I have used it in the past I have kept the same conversation going so it remembers things we had already discussed. However, over time (this is the free versions as well) the chat becomes so long that you begin to notice it forgets important details and then you spend time correcting it until you realize you have probably reached a point where you know your org so much better than the AI does.
From what I gather in the thread you are feeding it a fresh .csv file every time you use it so maybe you don't have such issues, but that brings me back to how it is going to value things if it isn't pre-set? What happens if you put in the same .csv file twice as a test? is the information the same both times?
I write none of this to be critical of your creation because it looks really cool. This is more about my frustrations in using AI in this role and am really wondering if you had figured some things out that I have yet to discover myself. Thanks.
|
Great thoughts and reply. In this particular use of AI, it maintains contextual awareness of the data presented within the app. CSV exports expand each time they are exported instead of being standalone, if that makes sense. So, the app and AI continuously see the growth of data providing a progression of the season, league, etc. With that said, it is just AI and is going to make mistakes.
As we keep building this and getting feedback from our own testing and everyone here, we will keep improving it. As of now, the logic for the AI chat is rather simple with little no guardrails. Fine-tuning it, pre-setting parameters, definitions, etc., are things that can and probably will be done, but as of now, I have had no issues asking it the set of questions I had and it being accurate.