How does it work
Last updated
Last updated
When a user types in a task, it passes through a series of checkpoints, such as language detection and moderation (to confirm if the command is hateful, etc.). This information is then sent to the AI, along with a specific engineered prompt (the "System"). We use GPT-4 to decrypt the tasks and send back the scripts and specific logic to our library.
For maximum performance, GPT-4 is recommended over GPT-3.5. Users can make up to 10 requests using Our API key, after which the system downgrades to GPT-3.5. Users can insert their API key if they wish to continue using GPT-4 and select the model they need. This downgrade is implemented because using GPT-4 is expensive.
We have implemented multiple checkpoints to ensure that the received logic is C# and Unity acceptable. These checkpoints are fed back to the AI model if any problem occurs, and the AI is capable of reiterating itself and fixing the identified errors. The AI can go through an infinite number of iterations to fix problems, as it reasons back and forth with the errors or any issues. However, for the sake of resource consumption, We have limited the AI to 3 iterations and a short memory span. These checkpoints help the system stay intact and avoid crashes at runtime, as certain logical inconsistencies can throw exceptions and halt a WebGL build.
Users can type the word "Abort" at any time to terminate a specific task.
Prompt: Create a mini solar system with all the planets. Color each planet based on its actual color. Include all the planets under the sun, let them rotate around it and themselves.
Logic Flow: The system sends the adjusted data to the AI for interpretation. Once it receives the information, it performs its calculations etc., and sends it back to us, where we decrypt the data and start performing what was requested. The AI also debugs itself and stores the information for future reference as it creates more tasks. As mentioned earlier, the AI is still under short-term memory constraints; therefore, we erase its memory of the tasks after a designated amount of successful tasks.
As You can see in the image and video below, it generated a sun, positioned the planets at a specific distance, gave them specific rotational and orbital speeds.
