Hey guys!
Yesterday i became aware that you can now create “gems” in gemini. I uploaded a copy of my Ironsworn - Sundered Islands rulebook, and tested it a little. It functions okay with writing detailed descriptions of the outcome of the moves, and offers helpful ideas. It’s wonderful at looking up tables and rules if needed.
Where it was a little wonky - and i didn’t expect this - was dice rolling and math. When starting out, it misunderstood how challenge dice and rolling for moves should work, but after that, it kept screwing up adding modifiers to dice, or said “strong hit” for a roll that was a weak hit.
All in all though, i was fairly satisfied with it. I enjoyed the narrative descriptions, and it was nice to focus on where i wanted the story to go and have “somebody” to throw ideas at.
Where it was a little wonky - and i didn’t expect this - was dice rolling and math.
I’m not really surprised - these are LLMs = Large Language Models. They are made to remix natural language sources, they can walk like a duck and quack like a duck. But the things that are really hard to get right with them are math, logic & reasoning, and facts. E.g. https://www.forbes.com/sites/johnwerner/2024/10/07/ai-is-usually-bad-at-math-heres-what-will-happen-if-it-gets-better/
AI image generators are even worse, e.g. https://gpt.gekko.de/ai-numeracy-text-image-models/
That is correct. The reason I was surprised was more because I would have thought that the training that go into such a large model as gemini would be on top of that.
At any rate, i’m glad you read my little post :)