The point is, while the models are inaccurate, the data (and there is a lot of data) indicate that warming is happening. This is shown not merely in temperature measurements; but in ice coverage changes, both in the Arctic Ocean and at mountaintops; and in biological change: increased growing seasons, earlier spring flowering, earlier bird migrations (and in some cases increased overwintering bird populations), increased ranges of species, including pests; not to mention permafrost melting and increased droughts and other severe weather anomalies in many areas.
I'm more interested in the theoretical question here, though. I'm unconvinced that self-similarity in chaotic systems means that since weather is a chaotic system, climate is itself a chaotic system that's unpredictable in the long term. One way of thinking about chaos is that at some point in the future, a system's position in phase space will become wholly uncorrelated with the system's initial position.
So for weather, tomorrow's weather is correlated with today's weather, but weather two weeks from now (or whenever) is wholly uncorrelated. But with climate you do end up with persistent correlations, year on year, decade on decade, millenia on millenia. Temperate regions show (anti-)correlations of deviation from annual average temperature between 180 day periods, and daily solar energy impingement is correlated with positive deviation from global mean temperature.
I've always made a mental analogy between that and a gas (the n-body problem): even though the positions of gas molecules rapidly become uncorrelated from their starting position, certain statistical properties--e.g., the pressure the gas exerts on the walls of its container--are highly correlated from one moment to another. Not just initially but into the indefinite future.
I'm open to arguments that that analogy is flawed and could see some assumptions that I'm not 100% confident in, but I don't see any obvious flaws. What's going wrong here, in your view?