AI AGENT/LLM: What did you teach AI to make them effective?
Smartsheet's AI is great! However, this question is for those of you who are also using LLM's such as ChatGPT, Gemini, Perplexity, Claude, etc. for creating formulas, asking best practices, analyzing data, and just generally getting answers to your questions more effectively and accurately.
Questions:
- What did you teach your AI Agent/LLM so that they effectively help you answer your questions?
- Which LLML do you use?
If you don't know the prompt(s) you used to train them, just do your best to explain how you did it :)
Maybe by sharing we can all improve on each other's knowledge to make it even better?
I'd love to see your responses!
-Neil
Comments
-
I've used ChatGPT 4o for some idea generation, ultimately I found that I normally matched up pretty well with what it recommended, but it certainly helped with speed.
- I asked it to act as an expert in Smartsheet and Smartsheet formulas, as well as acting as someone in my role/industry. I then primed it with stating the end objective/purpose of the sheets.
- Used ChatGPT 4o.
I wanted a sheet that would grade a finance application across 20 differently weighted criteria. I asked it to send me an Excel file and tell me the necessary formulas that I could paste into the sheet, then imported into Smartsheet. That saved time creating the 40+ columns individually and writing out any formulas. Then, I took the formulas that it generated for the weighting and plugged them in (had to make a few tweaks to get them to be proper column formulas).
Overall worked fairly well, pretty basic example but it saved a lot of time with structuring and naming the columns with this method.
I'm curious what others have done!
-
I use Gemini pretty regularly to do a number of things, both personally and professionally. It can be really finicky with complex Smartsheet formulas, but it's better than Smartsheet's internal AI.
I often use it to draft things like LinkedIn posts, letters, etc. I expect that it'll get me 90% of the way there and then I'll have to do some fine-tuning and that's generally the case. However, whenever I do the fine-tuning, I also paste that back in and tell it "This is the final version I'm using. Please learn from this." Doing that has helped it get better and better at understanding how I word things, the tone I use for different types of writing, etc. and that has cut down quite a bit on how much I have to tweak.
One thing about Gemini that's different from ChatGPT is that, from one thread to another, it doesn't keep a universal memory. I therefore have specific threads going for specific use-cases and just continue to come back to them as needed.
