That XKCD concatenated data from different studies with different temporal resolution, so it's very misleading. Past data came from Marcott et. all, who said:

“the paleotemperature records used in our study have a temporal resolution of about 120 years on average.”

“no temperature variability is preserved in our reconstruction at cycles shorter than 300 years, 50% is preserved at 1,000-year time scales, and nearly all is preserved at 2,000-year periods and longer.”

Recent data came from yearly measurements, so preserves variability on a yearly basis.

In other words, the way past data was determined, sudden changes would be hidden, like the temperature spike that it shows in the 20th century.

Do you ever wonder why that comic only goes back 22,000 years? If it had gone back just a little more, say 25,000 years (a rounder number), it would be much more complicated to explain what happened.

If it had gone back 400,000 years, which is basically nothing in geological scales, it would bring up some really interesting questions.