My Google feed has been loaded with ChatGTP and large language models (LLM) AI articles for the last month. Finally, we got some longer posts on the technology on DailyKos this week (sorry to not provide the links). I have primarily been interested in the programming and content generation elements having spent some years on a side project developing a web-based system. I have been somewhat dubious about AI prospects to build large complex systems but that may be very possible. At a minimum, a directed effort of code generation within a properly structured plan for the software could obviously be very powerful. Over last few years there have been many machine learning products developed that demonstrate the ability of the approach, given a proper framework, to complete targeted tasks with uncanny efficiency. I would stress “proper framework” and an understanding that on a case-by-case basis algorithms can still it wrong. I finally had time to experiment with something benign.
Yesterday afternoon, I spent no more than 20 minutes with Microsoft Bing’s playing with its new Chatbot (a far shorter time than it took me to assemble this post). My family likes to play group games and I had thought something that involved play acting in a soap opera context was a fun idea. This was my prompt for Bing’s AI:
Can you help me design a game that would consist of cards with short soap opera plot ideas. The cards would might having fun, crazy, or potentially lurid details. All players would assume some name for the game. Players would always have 3 to 5 cards to play at any one time. Some example cards might "I was looking through player 1's closet recently and discovered they had a baby" or "I got a call from a stranger recently and they told me player 2's used to go by a different name." The concept of the game is to let people pretend they are part of a soap actor and play act. I would like to try to get may be 50 cards to experiment with.
The bot gave me a bulleted list of content of a game based on this prompt that were roughly in line with what I was thinking even ignoring “would might having”. It generated the rules for game for me:
Rules:
- The game can be played by 3 to 10 players. Each player chooses a name for themselves and writes it on a card.
- The cards with soap opera plot ideas are shuffled and placed face down in a pile. Each player draws 3 cards from the pile and keeps them secret from the other players.
- The game starts with one player being the narrator. The narrator picks one of their cards and reads it aloud, replacing the player numbers with the names of the players. For example, if the card says “Player 1 is a clone of player 2 who escaped from a secret laboratory”, the narrator might say “Alice is a clone of Bob who escaped from a secret laboratory”.
- The other players then have to act out the scene based on the card, using their imagination and creativity. They can add dialogue, actions, emotions, and twists to the story. The narrator can also join in the scene as a character or as a narrator.
- The scene ends when everyone agrees that it has reached a satisfying conclusion or when it becomes too absurd or boring. The narrator then passes their card to the player on their left, who becomes the new narrator. The old narrator draws a new card from the pile to replace the one they used.
- The game continues until all the cards are used or until everyone decides to stop. There is no winner or loser in the game, only fun and drama.
These are not exactly what I had in mind but is a pretty good starting point.
Next, I asked if it could generate some example player names. I said it would be good if these could be funny and said each player should have a profession. It generated these example names:
- Blaze Storm, a firefighter
- Crystal Diamond, a jewelry designer
- Rex Savage, a lawyer
- Destiny Love, a romance novelist
- Lance Steel, a spy
- Angelica Divine, a nun
- Max Power, a politician
- Scarlett Rose, a model
A button offered to generate (and it did generate) names about 8 names at a time. Finally, I asked it to generate soap opera plots to go on the cards. Here are just some of the examples:
- Player 3 has a secret twin who is plotting to take over their life.
- Player 4 is actually the long-lost child of player 1 and player 2.
- Player 5 has been having an affair with player 1’s spouse for years.
- Player 6 is a spy who is undercover as a doctor in the hospital where player 2 works.
- Player 7 is pregnant with player 4’s baby, but player 4 doesn’t know it yet.
- Player 8 is a serial killer who targets people who look like player 3.
- Player 9 is a psychic who can see the future and knows what will happen to everyone.
- Player 10 is a vampire who has been hiding their true nature from everyone.
- Player 1 has a rare disease that can only be cured by a blood transfusion from player 2.
- Player 2 is a secret millionaire who inherited a fortune from a distant relative.
These are generic but aligned with my original prompt. Mostly likely these could have been much better by providing more framework for the prompts (e.g. lets make this game center around a family of oil magnates in Texas). These prompts also included multiple plot lines involving vampires when I requested more items, but if we were to play this game, I would have just removed some of the lines and added custom ones.
Too be honest, this was fun to do, and I feel it was pretty amazing that the technology could churn out this much with so little input. If I spent even a bit more time providing more framework (better prompts) and editing the output, it could be a fun activity for the family.
Anyway, I thought I would share this as example in case readers wanted to try something less serious than trying to get the bot to say crazy things. This technology seems likely to be semi-transformative in that it will further accelerate the ability to generate content (which would need fact checking and editing). It may also generate some large efficiency increases in areas like software development. It may be rife with intellectual property issues and other legal issues. That stuff is beyond me.
One of my beliefs about these chatbots and LLMs is that they will reflect the content of the prompts and ultimately the users’ beliefs and intents. That will be the primary source of good and bad in this technology more so than a runaway AI (but perhaps that will come to be as well).