We asked GPT-3 to help us choose a game for our company retreat. Here’s what happened!
GPT-3 team building activity
At Edge Analytics, we’re using GPT-3 and other cutting-edge NLP technologies to create end-to-end solutions. We have helped multiple companies build solutions that leverage GPT-3 (check out our recent work with inVibe using GPT-3 to improve the market research process), and our resident expert, Alex Browne, has put together a series of helpful articles on effectively utilizing GPT-3. So, when we wanted to choose a game for our company retreat, we decided to just ask GPT-3.
What is GPT-3?
GPT-3 was created by OpenAI and at the time of writing is the largest model of its kind, consisting of over 175 billion parameters. It was pre-trained using one of the largest text corpuses ever, consisting of approximately 2 trillion characters, which includes a significant chunk of all text available on the internet.
GPT-3 uses a text-based interface. It accepts a sequence of text (i.e., the “prompt”) as an input and outputs a sequence of text that it predicts should come next (i.e., the “prediction” or “completion”). Through this surprisingly simple interface, GPT-3 is able to produce impressive results. The trick is designing the right prompt to extract the right knowledge encoded within GPT-3.
Read more about using GPT-3 in our series on “Getting the Most out of GPT-3”.
“GPT-3 give us a game”
We split into two teams (healthy competition never hurt anyone, right?) and set a timer for 15 minutes. We had to figure out how to prompt GPT-3 to give us the best game. At the end of the 15 minutes, we would pitch our favorite GPT-3 response to the other team. (Full disclosure, our resident expert on GPT-3, Alex Browne, was not in attendance)
There are a number of ways to prompt GPT-3, and the way in which you prompt GPT-3 can drastically alter the end result. Here is a compilation of our favorite prompt-response pairs:
We started out simple…
Prompt: “Here are step by step instructions on how to play a fun game with multiple people”
Response:
- Think of a name for the game
- Take turns saying the name of the game in random order
- When you say the name of the game, throw a softball or a bean bag at the person who just said the name of the game. It may be helpful to explain the name of the game after you have said it.
- The person who was hit has to say the name of the game again
- Play keep away with the name of the game. Throw it to another person.
This is an advanced version of dodgeball where you are dodging words in addition to bean bags. However, we seem to have an infinite loop. How do you lose? How do you win? What is up with this obsession with the name for the game? Rating: 8/10
Okay, let’s take it up a notch with some comparison keywords:
Prompt: “Here are step by step instructions on how to play a fun game with multiple people that is more fun than Laser Tag”
Response:
This game requires one laser pointer. Get multiple laser pointers if you have more than 10 people playing.
The target is counted down on 3. BOOM!! Divide into teams.
Everyone needs to cover their eyes when the laser is shot unless you are the one with the laser pointer. If you are, the WEAPON should be covertly aimed towards one of the teammates. The first person bombed with the laser should count to 5 and then see who the laser is refracted off of. Team with the least laser hits wins.
So, laser tag? I think you missed the prompt here, GPT-3. However, I am giving bonus points for the safety briefing to cover your eyes. Safety first! Rating: 2/10
One of the teams took a different approach, prompting GPT-3 with the start of a game.
Prompt: Using only one hand at all times, take one Jenga block each turn from any tower level other than the one below an incomplete top level. Place the block at the topmost level of the tower. On the count of three (usually accompanied by a chant of “Rock, Paper, Scissors”), each time raising one hand in a fist and swinging it down on the count, opposing players “throw” their selection into the middle.
Response: The player whose selection comes down last (i.e. is “Rock”) removes one or more blocks from the top level of the tower, then the player whose selection came down second (i.e. is “Paper”) removes one or more blocks from the level just below that. The player whose selection came down first (i.e. is “Scissors”) removes one or more blocks from any level of their choosing. The number of blocks to be removed is determined by the number of players. At the end of each round, score the number of blocks remaining in each player’s tower. The player with the most blocks remaining wins.
The following variants are in common use:
In his book “The Dice Man”, Luke Rhinehart describes a variation of Jenga called “Castle Fang” (named after the castle in the illustrated book), which involves the use of a bottle of strong spirits and a dagger. The game is played identically to the original, but after a tower is built, the loser of the game may drink from the bottle and stab the player who won the game with the dagger.
A variation described in the book “The Worst-Case Scenario Survival Handbook” involves three people, each with three blocks. One person takes a tower apart in the usual way. Each of the other two players tries to knock down the tower so that it falls on the other player’s blocks.
At first it appears to be a fun combo of Jenga and Rock, Paper, Scissors. But then you get to the variants. Things get…umm…dangerous:
The game is played identically to the original, but after a tower is built, the loser of the game may drink from the bottle and stab the player who won the game with the dagger.
Overall, the rules are clear and the objective of the game is stated. GPT-3 gave us the family friendly version as well as two variants that are a bit more exciting. Rating: 9/10
So what was the result?
Asking GPT-3 for a game for our company retreat was a fun team-building activity in itself. We highly recommend it as a team building exercise for other startups. By adding the team pitches, it added another layer of hilarity (while also practicing our pitching skills!).
In the end, we decided to play mini golf instead of playing one of the GPT-3 suggested games:
Edge Analytics has helped multiple companies build solutions that leverage GPT-3. More broadly, we specialize in data science, machine learning, and algorithm development both on the edge and in the cloud. We provide end-to-end support throughout a product’s lifecycle, from quick exploratory prototypes to production-level AI/ML algorithms. We partner with our clients, who range from Fortune 500 companies to innovative startups, to turn their ideas into reality. Have a hard problem in mind? Get in touch at info@edgeanalytics.io.
