LAB31

Play a virtual round of golf against Tiger Woods at Pebble Beach.

Try the GPT here (Requires GPT+ Subscription)

This project was months in the making, though it wasn’t until we dove into the 31 GPTs project that we figured out how to make it work the way we wanted it to.

The game is a text-based game of golf, where the GPT will describe the shot ahead of you, you describe the shot you’d like to take, and the GPT will describe the outcome of the shot. If you think this sounds like something ChatGPT should probably be able to handle, you would be correct. However, we both have a tendency of being overly ambitious with what we try to make this AI thing do, to the point where it almost kills us… and this is no exception.

This game has the following unique functionality, which made it a delightful pain in the ass to build:

1. It knows all 18 holes of Pebble Beach golf course inside-out, including distances, hazards, green slopes, and the views of the ocean.
2. It delivers shot outcomes relative to the person’s handicap. It turns out chat GPT is not naturally good at making people bad at golf. We forced it’s hand by making it roll dice in the backend. A 36 handicapper has a 36% chance of duffing, chunking or shanking a shot. A 10 handicapper has a 10% chance.
3. Every shot takes all key variables into consideration: Wind speed, club selection, shot type, lie, custom inputs (if you ask it to do something wacky), and random factors. For each variable it rolls a dice. It adds up all the dice to create the Total Difficulty Score (TDS). The higher the TDS, the worse the shot outcome.
4. It keeps track of the score using a python environment.
5. At any point, you can ask it to visualize the lie of your ball and it will spit out an image using Dall-E.

These days we’re on the open source wagon, so we’ll save you the hassle of trying to hack this GPT. Here are the game’s instructions, featuring all of the dice roll logic.

Let’s see this thing in action:

Golf is hard.