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:

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.