MemT-1.0: AI Memory INSIDE the Model Is CRAZY (I Tested the Idea)

MemT-1.0 puts an AI's memory inside the model. The authors claim about 66x less persistent storage than a full KV cache at 128K history. I could not run MemT on my 8 GB laptop GPU, so I built a toy test of three memory styles: full chat history, file retrieval, and a fixed-size compressed state. This is NOT MemT's architecture. Chapters 0:00 The 66x claim, and the problem 0:53 What MemT actually is 1:56 Three ways to give an AI a memory 3:49 The claims, and who measured them 5:01 Can I run it? (no, here's the proof) 5:40 My toy experiment: how it works 7:30 Results: full history, retrieval, fixed-size state 10:21 What it means for you 11:42 Links and question Links Repo: https://github.com/MemTensor/MemT-1.0 Paper: https://arxiv.org/abs/2610.26825 RunPod: https://get.runpod.io/pe48 Notes: the repo software is PolyForm Noncommercial; model weights terms are not published yet. Cache sizes in my toy test use a stand-in model config, not MemT's. Parts of this video use an AI-generated avatar of me. Get GPUs Runpod: https://get.runpod.io/pe48 Get Hostinger: https://www.hostg.xyz/SHJiu Coupon Code "PROMPT" 10% discount 🔗 Connect with me: Email → [email protected] Ko-fi → https://ko-fi.com/promptengineer Patreon → https://www.patreon.com/PromptEngineer975 Book a Call → https://calendly.com/prompt-engineer48/call GitHub → https://github.com/PromptEngineer48 Twitter/X → https://twitter.com/prompt48 Website → https://promptengineer48.com/