So, let's trace the HadCRUT4 data over his...
I don't know what date he pulled the 2016 data from, as that's his end-point, so I referenced the latest HadCRUT4 chart I could find (February).
It should be noted that like 1998, mid-2015 to mid-2016 was an El Niño period, which is why you have that big uptick at the end of the graph. 2015 was supposedly the hottest year on record, and the first half of 2016 was hotter still... but the back half of 2016 will be cooler... so when all is said and done, the end of my trace (which is higher than xkcd's) will be slightly lower.
It seems probable that the xkcd graph is smoothing out both El Niño events. It traces a pretty middling line between colder years and hotter years.
Note that he helpfully says near the top of the graph that a single year or single decade spike could be smoothed out by the prehistoric data, but not a multi-decade trend (well... that's what it seems he's saying).
Warning. Long image.
Whoops... was that too late?
Thanks for doing that. It's perfect. See how in the black line there is a trend from 2000 to 2016, and see how that trend is entirely based on the final data point? It should flatten like the yellow line does through that period. I'll draw something up to illustrate my point since you guys think I'm smoking crack.
Edit:
Alright here it is. You guys are making me go to ridiculous lengths to demonstrate something which ought to be easily observable by eye, but here you go. I'm on a weird old computer doing this with a tiny screen, and I had to install packages in linux to be able to edit it, but that's what I'm willing to do to answer the call. Since you guys obviously think I'm a moron, here's what I'm saying (as a disclaimer, I probably screwed something up, I'm doing this as fast as possible):
The red is the XKCD line based on pixel analysis of where his thick black smudge crosses 2000 and 2016. The black ticks I put in to represent 0.1 degree increments and where 2016 is. I put red error bars representing the outside of where his black line fades out in the image. His black line crosses between 0.29 and 0.42 degrees at 2000 and crosses between 0.64 and 0.83 degrees at 2016. If you go center-to center that's 0.355 to 0.735 for a difference of 0.38 (pretty close to what
@TRGTspecialist said). If your eye goes outside to outside (which is undoubtedly what mine did) you get 0.54 degrees. So I'm not some lunatic. I was looking at blank space on one side of his line and blank space on the other side, which, admittedly, is not fair.
So he jumps 0.38 from 2000 to 2016 from center to center. He does so by starting below the data at 2000 and ending close to 2015 on the high end, but more like centered at 2014. The green line is where he should be, with a nearly flat (0.1 degree increase) over 14 years followed by a big jump in 2015. Now when you exaggerate the growth over 16 years by several times (and it helps to downplay the growth prior to 2000 in this case), you can draw a much more aggressive line when you extrapolate that out 80 years.
Bottom line, I'm right, XKCD overplayed. You guys must, at this point, very clearly see my point... which has now been beyond thoroughly beaten to death.