This is exactly what I started postulating on about 5 years ago… that eventually, transformers working through a REPL with access to a vector-store could likely

This is exactly what I started postulating on about 5 years ago… that eventually, transformers working through a REPL with access to a vector-store could likely lead to AGI. Of course, I didn’t predict the LLM/multimodal explosion, but I’ve been thinking along these same lines for a while now. My current direction is a multiagent MOE working into a single REPL with a supervisory transformer that manifests intent through the management of agent delegation and response filtering /regeneration to stay on context.