I do and and don't, but you are bringing up some good points. Unfortunately you're also demonstrating point two. An unfortunate limitation to my above statement is that it doesn't work so well with strangers on the internet. So instead let me talk about the scientific process, as a scientist myself.

The above isn't how it works. But if it was, it might not even be wrong. Let's start with that. First off, why were they creating 1000 or even 1000's of models? You're on HN so I assume you program. In a big program how many times do you have to rewrite things before you get them working perfectly? Then how are these models accuracy determined? Well, with predictive power. The current weather models have high predictive power to 3 days (the European model has about 5). After that it is clear that error greatly increases. But that's weather, not climate, and I want to differentiate the two here. Climate modeling really is different from weather modeling. We have a lot of data on it, so it is more performing analysis to figure out the cause for changes. So let's ask ourselves how can we check the prediction power of our climate model? Well if our analysis is correct we can go about it two ways, or really check with both. We can use historic data (if you've ever looked into stock analysis, this is a common practice there). So we feed in our key variables and ask what temperature we get back. Does it match what we measured? And how well? We can also use it on new data, because as each day goes by we get more data.

I'll stop there, as I don't want to write an entire paper. But a key part to science is its predictive power. If we don't have predictive power it isn't science, plane and simple.

p.s. If you just plot the raw data (not modified for methods in which that data was gathered) you see a trend much worse than what current climate models predict.

p.s.s Current models have held up in predictions over the past few decades. But there is always room for improvement.