Fable Model Shows AGI-Like Sigmoid Learning on EdgeBench
Source Posts on X
Cluster sources
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTexTECH
Some question my "Fable is probably AGI" take. The valid pushback: "so what's new? Why not Opus 4.8/4.5?" I think the difference is in what EdgeBench measures. They all saturate. Fable is "AGI" enough to reduce ANY state filling its context into the beginning of another sigmoid. https://x.com/teortaxesTex/status/2074471487579332707/photo/1 https://twitter.com/zktrivo/status/2074454063589347794
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTexTECH
Or at least, that's the goal. What I can say for sure is that Fable/Mythos was trained with the ambition to *maybe* get AGI, an intelligence capable of unbounded self-improvement, and it's been expressed in specific, serious design commitments.