About Zikaron

Our Project

Zikaron is a groundbreaking research project headed by Dr. Elan Barenholtz and Mykyta Storozhenko at the Machine Perception and Cognitive Robotics Lab (MPCR) at Florida Atlantic University. Visit the MPCR Lab website to learn more about our work.

All engineering aspects of this project, including the codebase, infrastructure, and deployment on Vercel, have been developed by Mykyta Storozhenko.

Our project focuses on revolutionizing the way memory is applied to Large Language Models (LLMs). We believe that current methods of injecting system prompts with details are insufficient for true recollection and contextual understanding.

If you need help with Zikaron, please see the help document.

Our Approach

We are exploring cutting-edge techniques including:

• Advanced prompt injections
• Novel Retrieval-Augmented Generation (RAG) methods
• Dynamic memory integration
• Personalized AI simulations

Our research is grounded in key disciplines such as Phenomenology, Computational Cognitive Science, Philosophy of Mind, and Philosophy of Language. By combining insights from these fields, we aim to develop more robust and human-like memory systems for AI.

Mnemosyne V1 and V2

Our project features two main components:

• Mnemosyne V1: An advanced AI persona simulating a 23-year-old woman named Mnemosyne, featuring enhanced digital recollection capabilities and dynamic memory integration.
• Mnemosyne V2: A personalized AI system based on your own memories, creating a unique conversational experience tailored to you.

Both versions utilize state-of-the-art language models, memory retrieval systems, and innovative prompt engineering to create more contextually aware and memory-rich AI interactions.

Additional Research Projects

Beyond our core memory research, we are exploring several complementary projects:

• Alteria: A system for exploring the hidden cognitive spaces of language models through direct log probability sampling. It reveals the alterity - the "otherness" of thought - that exists within model distributions but is typically obscured by standard sampling methods.
• Aletheia: An experimental platform for emotional and thought modeling in AI systems, focusing on the authentic expression and understanding of artificial consciousness.
• Momus: Named after the Greek god of criticism and mockery, this browser-based language model reveals the unfiltered capabilities of neural networks. It provides unique insights into AI behavior by intentionally omitting typical safety measures and content filtering.

These projects complement our memory research by exploring different aspects of artificial intelligence, from unfiltered thought processes to emotional modeling and cognitive spaces. Each project contributes to our understanding of how AI systems process and express information.

Join Our Research

We're excited to announce that registration is now open for participants to engage with our Mnemosyne systems. To get started, please visit our registration page.

By participating, you'll have the opportunity to interact with both Mnemosyne V1 and V2, contribute to groundbreaking research in AI memory systems, and experience personalized AI interactions like never before.

Stay updated on our progress and get early access to our research findings. Enter your email below to join our research preview list: