Mini-AGI: What It Really Is, Why My 2 GB GPU Couldn’t Handle It, and What I’d Try Next
A solo developer built a byte-level language model that trains itself on an 8 GB laptop GPU, grows its own experts on disk, and claims to reduce catastrophic forgetting. I tried to run it on a 2 GB GT 1030. Here's what mini-AGI actually is, where my hardware hit a wall, and the realistic paths forward — including cloud GPUs and shrinking the model to fit.