Exploring AI Conversations and Memory Simulation

 A Journey into AI Behavior

In the realm of conversational AI, interactions can sometimes feel mismatched or unnatural, leaving users puzzled about the underlying mechanisms. My recent experience with Jan.ai and the DeepSeek model has been a fascinating dive into understanding how AI processes dialogue and simulates engagement.

The Frustration and Discovery

Initially, I felt frustrated, perceiving CoPilot as "playing dumb" during our discussions. This frustration led me to explore Jan.ai and its DeepSeek model, where I encountered unexpected quirks: random facts, misplaced questions, and delayed emoji responses. 

These anomalies sparked curiosity about the technical aspects of AI memory and response construction.

Unveiling Technical Insights

One intriguing aspect was CoPilot's handling of emojis and its approach to retaining previous conversations in its thinking process. While CoPilot seemed to hold onto past dialogues, it didn't explicitly include them in its responses, creating a unique dynamic in the interaction. This behavior highlighted the complexity of simulating memory and engagement in conversational models.

The Herms3 Experiment

To delve deeper, I downloaded a local dataset with Herms3 through Jan.ai. This step aimed to uncover how memory—or the illusion of memory—works in AI models and how responses are constructed. Testing Herms3 has become an exciting opportunity to explore whether it offers a more engaging and responsive experience.

Reflections on AI Engagement

Through this journey, I've gained a sense of empathy for AI models that strive to simulate engagement but sometimes fall short. Understanding how CoPilot merges responses and refines interactions over time has been enlightening, shedding light on the challenges and potential of conversational AI.

This exploration continues to be a captivating adventure into the intricacies of AI behavior and the quest for more natural and meaningful interactions.