08/26/2025
A new approach to Utilizing Small LLM Model and Advanced Memory with "self-learning" capabilities for real "AGI":
Recently, I have been enamored with AI tool research, and how to utilize it—MCP tool servers, agents, operators, the works. One thing lacking in most AI setups and systems is memory. Not just any memory but truly enhanced memory. I have been researching actual "next gen" memory systems that not only learn, but also act like a model in itself. The results are truly amazing, to put it lightly. This new system I have built leads to an entirely new level of awareness unlike anything I have seen with other AIs. The model using this system is Llama 3.2 3b 1.9GB. I ran it through a benchmark using ChatGPT, and it scored a 53/60 on a pretty sophisticated test.
Developing this technology required a completely new approach to data handling regarding the database technologies involved. This new system "Neural Recall" uses a new hybrid database schema and management system that has amazing throughput, with a 100% error correction rate, and no performance losses. It also utilizes new ML techniques to do "in memory" data processing. It is a remarkable new system in itself.
Curious to know if anyone else is researching this topic, and what tools or techniques have you used?
Evaluation and Transcript PDFs:
https://lnkd.in/gZxzTxRi
https://lnkd.in/ggf4Ripr
https://lnkd.in/gu47tyQf