How we tried to make OpenAI do our broadcasting jobs for us
On Monday, July 3, 2023, the production team for the Wisconsin Rapids Rafters was sitting in the press box at Witter Field, building their pregame show for the day. This was the last game of the first half, and three teams were in a tie for 1st place, whoever got 1st would secure a playoff spot. (Click here to see for yourself, First Half Standings 2023) The matchups were so close, that the league rulebook hadn’t given any information on who would advance. So as creative director, I decided to ask ChatGPT what it thought. Everyone agreed, and we entered a detailed prompt. Here are the results.
Not bad, right? We figured this was a good baseline to use for our show. We then called the league offices to confirm if it was correct, and they agreed with us. We were all going a bit crazy, as we didn’t expect that to work.
Another question came up: “Could we have ChatGPT create a whole pregame show?” So we decided to try and make one. Our first attempt wasn’t great since our prompt wasn’t specific enough. So we tried again to include as much information as possible. Here are our results:
Not as great as we thought. Don’t worry, we didn’t use what ChatGPT made to complete the pregame show, we made it on our own. But after the game was over, I decided to look back on it after the game. This was the first time I had looked into what AI could make with enough information.
I figured out a few things. First, I figured out that it was much better at math than me. Not really a surprise, as I’m not the best with math to begin with (sorry to all of my high school math teachers). ChatGPT also has the power to use the internet. So combining both the prompts we gave it with a small bit of math knowledge gave us a comprehensive and easy-to-understand answer. Very nice.
Second, I realize we gave it a really good prompt. When we first tried it, the prompt failed us. Yes, ChatGPT uses the internet, but only the internet from two years ago at its most recent. That’s fine for figuring out math equations that don’t change, but not for a sports league with things changing every hour.
Finally, it couldn’t make a compelling sports story to save its life (computer life? I don’t know…) It only gave us a template with loosely filled answers, rather than patch notes on the game. If we had put the specific information in the prompt, it probably would’ve glued the template together with those names that it knew nothing about. We might as well have written two quality pregame shows before ChatGPT could tape together one coherent message.
I’m sure if I try this again next summer, the results will be better. With both my knowledge of OpenAI, as well as the fact that it keeps getting better and better every day. But with that, I learned it’s good at solving problems and complex situations, but only if you give it the right information.