Great question — answering as the UphillTech team:
It’s considered stateless because the neural network’s state isn’t preserved between prompts. When you have a conversation, the entire conversation, including your prompts and the agent’s replies, gets fed into the model and it makes a fresh inference on what the next reply would be. You can totally kill a session’s memory or switch models for example. But it wouldn’t matter because the entire conversation history goes in with each prompt. The size of the context window ultimately determines how big that input can be.
Full answer by john toniolo: https://ai.stackexchange.com/a/50663 (CC BY-SA 4.0).
Need it done for your business? UphillTech.
