How good is the weather forecast to 2100?

The UN's climate change wonks, the IPCC, have over their nineteen-year life released three scaremongering reports (the fourth is released today) that have included various predictions of the future -- predictions that have varied with each report -- predictions that purport to know the weather forecast right up to 2100 -- predictions on the basis of which politicians are champing at the bit to "take action" to stop private action.

Now, since their long range weather forecast is so important, you would think that for after nearly twenty years of forecasts you would think that we might be able to review these predictions for accuracy, and it turns out we can. In the latest report there's a simple graph, pictured below, that measures the IPCC's three previous predictions against a "trend line" (in black) drawn between 'spots' that show the "global average surface temperature" for each year (with all the attendant reservations about that record).

Have a good look, and then perhaps you might care to answer these three questions:

1. Which of the three predictions do you think has been the most accurate? 1990, (shown in blue with a 'best estimate' and a suggested range)? 1996, in brown, or 2001, in blue-green (by which time they'd given up on a 'best estimate' and instead had just fired a shotgun at a whole range of predictions).
   
2. What does the level of accuracy for this period of time -- or lack thereof -- suggest for the IPCC's predictions over a whole century.
3. Given that none of us know the future, and this is all warmists have to go on, then how seriously should we take these predictions?

To put it simply, are these the sort of predictions that you'd put money on? Because you're being made to.