I think we are pretty far. I am not devaluing the o3 capability but going through actual dataset the definition of "handling novel tasks" is pretty limited. The curse of large context of llms is especially present engineering projects and does not appear it will not end up producing the plans of a bridge, or an industrial process. Sone of tasks with smaller contexts sure can be assisted, but you cant RAG or Agent a full solution for the foreseeable future. O3 adds capability towards agi, but in reality actual infinite context with less intelligence would be more disrupting at a shorter time if one was to choose.