Welcome to a very special edition of the show! Today, we will be playing a game called Wheel of Reason. This game is similar to the famous TV show that you all know and love, but with a twist. It is programmed in Python and freely available on my GitHub. You can download a copy for yourself and play along if you like. The instructions are included in the description.
The main file of the game is called ‘w o r pi’. You can easily run this file. The game revolves around a wheel and various options. The categories include places, phrases, things, people, and titles. You are welcome to change any of these variables and options to create your own unique experience.
The purpose of this game is to test the capabilities of new chat models in terms of reasoning. It requires the contestants to guess the missing letters and solve the puzzle. The chat models need to understand the question and make the best decisions to move forward. Of course, luck also plays a role, just like in real life.
In this special edition, I have provided each of the contestants or AI models with a prompt describing the game and how to play it. The models have responded properly and shown an understanding of the game. I will be playing the game in an accelerated mode, but you can pause it at any point.
I have observed that Google’s chat model consistently performs well as a contestant in the game. It provides recommendations and strategies on how to play the game. I did help Google’s model a bit more than I would have liked to, but I wanted to give it a fair chance.
On the other hand, I faced some token generation issues with Hugging Face’s Llama 2 and Cloud version 2 from Entropic. These models have more restrictions and limitations. However, Microsoft’s Bing AI model is fully available to the public. Although I had to use a more creative conversational style with Bing, it was able to play the game effectively.
In conclusion, Microsoft’s Bing AI model has performed impressively in the game. It has shown robust performance, although not flawless. It sometimes gets lost but overall performs well. It understands the game and provides answers that are obvious to a human contestant.
In the next edition of Wheel of Reason, Bing and OpenAI will go head-to-head to determine the finalist. Stay tuned for the exciting competition!
Thank you for watching. I am Ray Bernard, the OpenAI tutor. Please hit the subscribe button and see you next week!
