Yet there are those who claim the ability to predict weather 85 years out.
In fairness, predicting long term averages is easier than specific weather on a given day. Saying it will be cold in winter is a lot easier than saying "it will be 30 degrees on December 27".
However, that is based on a mental model we all carry in our heads that says it's always colder in winter. Saying it will be colder in winter conforms to our experiences. Reality ALWAYS matches the model. Yet when you build a model that does not match reality (i.e. it makes assumptions about future states that we've never seen before, like rapidly escalating temperatures over time) you are really out on a limb. There will be errors in those assumptions, in the assumptions themselves and in the behavior of those assumed conditions over time. Under the best of conditions you might be able to make some short term predictions but errors compound over time. The longer the timeframe you are projecting, the less and less accurate you are going to be. So the net result is that such long term models are academic curiosities with very little chance of being actually correct.
I'll give you an example. If you are operating a vessel at sea and you have no ability to use the stars or GPS to navigate then you will have to navigate inertially. That is, you will measure your speed and direction from your last known position and make an estimate of where you are based on that. But every time you mark your estimated position on the map, it will have some inaccuracy. If you know your business it will be pretty close but the sensor you use to measure speed won't fully account for undersea currents or other factors, for example. Your clock will also not be perfectly accurate so you have a slightly bad speed and slightly bad time and you use that to calculate distance traveled which is consequently somewhat inaccurate. All those compounded errors will show up as small errors in the position you calculate. You mark the map and you draw a circle around what you think of as the position and say "I think I'm here but it could be anywhere inside this circle really". The issue is that you will then have to use that inaccurate position as your starting point for your next position estimate. And you make a guess again (with all those errors again) and the the new guess of where you are has to have an even bigger circle of 'I could be anywhere in here' around it. Eventually, that circle becomes so large that you're just flat out lost and will never find your way home. Inertial navigation is something ships at sea do all the time and have for decades but even after decades of real world use and adjustment and refining of the error models and the calculation algorithms, unless they get good position fixes periodically to reset the errors, it all falls apart in a matter of days or weeks (for the absolute best systems in the world) and you are simply lost.
This process is exactly why long term climate models, in my opinion, are about 100% worthless. They make some big assumptions about, say, global concentrations of certain gasses over time, assumptions that may or may not come true. They they make certain assumptions of how those gasses will affect the climate, assumptions that also may or may not be true. And then they take many many other similar models, with inherent uncertainties in them, and estimate the overall impact on climate over weeks, months, years, centuries. They validate those models by saying "well, it was pretty accurate in its prediction of climate over a very short timeframe" and use that as justification that it will be just as good at predicting things over decades. But that's is flat stupid.