Living Weights experiment bakes what your model learns directly into the weights
Jasonio · x · 2026-11-17
ashxhart introduces Living Weights, an experimental project: conventional open models ship with frozen weights and forget you when the chat ends. Living Weights keeps learning between replies, saving new knowledge directly into the weights — surviving restarts and even machine transfers, with no KV stashing or memory files, all on your own hardware.
It's still experimental; the authors are iterating with petruspennanen.
More from Models
- InternLM's Sage-S2.1-Flash model trends on Hugging Face — internlm · 2026-11-17
- Uncensored DeepSeek-V4-Pro-Qwen3.5-4B 'heretic' merge trends on Hugging Face — xtools-at · 2026-11-17
- Uncensored DeepSeek-V4.1-Flash variant trends on Hugging Face — MoonMath-ai · 2026-11-17
- Benchmarking DeepSeek V4.1 Flash, GLM 5.3, Kimi K3, Mistral Large 4 on Finding a Real 0-Day RCE — matthew_d_green · 2026-11-17
- New MAI Playground UI leaks ahead of Microsoft Ignite on Nov 17, where new MAI models are expected — marlene_zw · 2026-11-17
- d1-Omni tested on 100 receipts: 98% date accuracy but only 52% on full addresses — mervenoyann · 2026-11-17