>> according to the best estimates is likely to be positive over the coming century  >>

Lets hope so, but clear sky days seem to be increasing.  We could do with much more rain.

Depends on where you are. This year and last, in New England, have been among the wettest on record. This year, so far, we've had a good 20 to 25 inches more rain than "normal", which is substantial! And we've also had many cloudy days. This past spring, for example, it rained on (I believe) 21 days in May and some of the days that didn't rain were cloudy. Luckily we did get some sun during the summer. But this gray, rainy spring seems to be the trend. And warmer winters, more rain, less snow. Then there are other places where rain is desperately needed.

This might be a good place to note the "cosmic rays cause clouds; clouds reflect the sun; we are going though fewer cosmic rays now, therefore fewer clouds, therefore more warming" theories of Svensmark et al. Aside from the general haggling about the accuracy of the model and whether there is any correlation at all, and the question of how a 14 million year cycle can cause rapid warming over a couple of decades, there is this basic problem: the current measurements show a greater effect for nighttime temperatures than daytime temperatures, and a greater effect at the poles, which are more covered in snow and ice than the rest of the earth and therefore already more reflective; the opposite of what any "clouds reflect the sun, less clouds means more cooling" theory would predict.

Ummm....  aren't you basically saying here "yeah, properly modeling clouds is about as important as it gets to understanding climate, and no, it's not currently done well at all."  That's not much of a response; more of a admission.  

What he's probably getting at is this:  I make a computer model, I have various measurements I take or assumptions I make that lead to uncertainty, some of the uncertainties are small, others large.  I carefully propagate my error mathematically through the calculation at the end and publish my results.  Let's say they boil down to, oh say something like, I dunno, a temperature.  The number is published as 5C +/- 2C.  Of the 2C worth of uncertainty about 80% is due to a single rather uncertain measurement or factor.  Thus I say "factor A accounts for the majority of uncertainty in this value".  This does NOT mean that my whole experiment or model in useless or invalid.

Of course the best test of a computer model is to match it against the real world.  This has been done through hindcasting.  You take known historical data on CO2, pollutants, etc... and run the GCM.  You then see how well the temperatures it produces matches what happened in the real world.  It turns out that the current GCMs are quite good at matching with known historical data in fact.  If the model clouds were behaving much differently than real ones this would not have been possible.

You also seemed to have missed the bit at the end where he told you that nothing in the past indicates that clouds will save us.  We've seen massive warming lately along with massive output of GHGs.  The clouds haven't saved us yet, why would anyone think they will suddenly start doing so?  And where were the clouds during the paleoclimactic periods that were very warm?  Napping on the job again.