iamtrask: AI attribution is closer to solved than most realize

iamtrask · x · 2026-09-19

iamtrask argues AI attribution is more solved than people realize, because the solutions don't self-identify as attribution: Shapley values handle inference, LoRA handles post-training, and hierarchy helps at pre-training. He predicts a future where everyone fine-tunes open-source models on private data and registers them on routers — yielding not AGI but "RGI" (routed general intelligence), a general network of narrow intelligences. That routing, he says, indirectly solves attribution: you get paid when others' agents want your model's unique knowledge.

Related event: iamtrask: AI attribution is closer to solved than you think(3 posts)→

Original post →

More from AGI Musings

AGI Musings channel →