I believe the bottleneck for AGI is our reliance on static, "pre-frozen" weights. If the logic is frozen during inference, it’s just a glorified lookup table. Real intelligence needs a substrate that rewires itself through experience.
As a passion project, I’m building an AI assistant that learns in real-time without any underlying LLM or external generative APIs.
Just a heads-up: This is a solo enthusiast project, and I'm still deep in the 'build-test-break' cycle. It’s very much a work in progress. I’m not claiming to have solved AGI — I’m just exploring a different path.
Show HN: Is the "frozen weights" paradigm the main bottleneck for AGI?
github.com