Predicting aggregate impacts of a warm climate decade into the future is far easier than predicting the short-term process of an unstable system hypersensitive to initial conditions.
If the model predicts that a glacier disappears in 2050 and it disappears in 2070 instead?
Or if the lack-of-model makes a lack-of-prediction about glacier disappearance, and then it disappears in 2070 by surprise?
Error accumulation makes sense to worry about if you're planning a voyage and worried that dead-reckoning navigation will leave you kilometres off course. But if you're on the voyage already, out on the open ocean, you can argue with the navigator's math but you still need to navigate -- you won't do better by just blindly insisting the boat never left the harbour.
Think of climate modeling as saying that for a given species of trees, half of them will be dead at a given latitude in 50-100 years, not pointing at a specific tree and predicting that it will die on July 7th, 2047.