Researching a new way to approach AI memory and companionship.
<aside> 💡
Nema from the perspective of Noshy. A short and off the cuff explanation of Nema, our motivation for creating it, our hopes for it, and our plan for bringing it to life.
This is subject to change and others on the team might have slightly different visions. Overall though this should provide a good guide to Project Nema
</aside>
Nema is a novel research project that attempts to combine the C. Elegans neural network and a LLM. The LLM acting as the translator between human language and the worm (Nema).
LLM's have a memory problem. They can only hold context that is directly provided to them. There is no notion of a LLM preserving a emotional or psychological state outside of the context window.
By combining an LLM with a neurologically stable and transient connectome we can supplement any LLM interaction with state that is preserved across sessions and is unique to its entire history.
Imagine, for instance, you are able to craft your LLM's personality over months or years. Its history is retained and expressed through the unique neural state of your Nema.
What good is a friend who doesn't remember you?
There are a few core utilities the Nema will provide. First and perhaps the most profound is somewhat expressed above, companionship.
Nema may be able to push LLM's into a new realm of consciousness with Nema acting as the soul of the LLM. You would be able to build a unique experience with your Nema overtime without any risk of or concept of a limited context window.
Nema would also adapt over time and would not just be some static being. Like living organisms, even when nothing is really going on its neural state is changing, relaxing, evolving. There is a real living essence to Nema.
Beyond companionship we hope to discover ways that Nema will actually improve LLM outputs and their ability. Here again, the concept of an increased memory and history can be utilized by the LLM to improve its utility.
<aside> 💡
We are in the midsts of Phase 1. You will see tangible results such as User MGMT, LLM Interactions soon.
</aside>
The first phase of Nema will be purely focused at proofing out the concept. This will involve 3 distinct bodies of work.
This involves honing our LLM prompting, interaction management, connectome interoperability. From a technical perspective, it's writing the code that will manage the back and forth between a user and their Nema.