Currently, Sydney operates as a single instance. However, as we expand her features, the growing complexity of the input context causes more frequent instances of hallucinations. This poses a challenge when implementing new features, as suggested in issues #13, #20, #6, and #23.
To address this issue, we propose implementing multiple instances of Sydney and GPT 3.5, each with its own dedicated context for executing specific functions. These AI instances would be invoked by the "main" Sydney instance. The main instance will consistently receive user messages and output a JSON object that can be processed in the code to determine which instance to call.
For instance, consider the following example:
User: Remind me to buy food at 13:00
Main GPT instance output:
{
"instace" : "reminder
}
inside the code:
if (GPTJson.intance == "reminder") {
//calls reminder instance
} else if (GPTJson.intance == "assistant") {
//calls assistant instance
}
This approach will allow us to create separate instances for various purposes, such as group-specific functionality, personal assistants, and any potential future additions. By effectively managing context and utilizing multiple instances, we can mitigate the issue of hallucinations and facilitate the smooth integration of new features into Sydney's framework.
Currently, Sydney operates as a single instance. However, as we expand her features, the growing complexity of the input context causes more frequent instances of hallucinations. This poses a challenge when implementing new features, as suggested in issues #13, #20, #6, and #23.
To address this issue, we propose implementing multiple instances of Sydney and GPT 3.5, each with its own dedicated context for executing specific functions. These AI instances would be invoked by the "main" Sydney instance. The main instance will consistently receive user messages and output a JSON object that can be processed in the code to determine which instance to call.
For instance, consider the following example:
User: Remind me to buy food at 13:00
Main GPT instance output:
inside the code:
This approach will allow us to create separate instances for various purposes, such as group-specific functionality, personal assistants, and any potential future additions. By effectively managing context and utilizing multiple instances, we can mitigate the issue of hallucinations and facilitate the smooth integration of new features into Sydney's framework.