Engram: A new type of AI

From the author: From my endless pursuit of getting LLM's to stop hallucinating I had a late night chat with gemini discussing what if's. What if we could use a vector db during training, what if we could allow for agentic reasoning at the point of training. What if the LLM could apply this reasoning between layers? I had my ai-agent code up a prototype and started training it only to be disappointed with word salad. I think I'm on to something here though- I recently came across a paper by deepseek, and it aligns with the ideas I had , I mean I even called my repo Engram - and this is an idea discussed in their paper coiencidence ? i think not haha But I need help figuring out training- Whats the best way to train this thing? Consider this less a how-to article, but more asking for help to keep pushing this thing forward- I think i'm on to something. * Word salad: * The code: https://github.com/kent-ai-dev/engram ⏺ How Engram Actually Thinks: A Deep Dive Into Its Reasoning Engine The