Individuating entities in a document using spaCy

[pythonski]

Hello! I have been reading the paper From Sparse to Dense: GPT-4 Summarization with Chain of Density Prompting. The basic idea in the paper is as follows: GPT-4 is at first asked to generate a summary of a document, using only one to three entities. In the following runs, the length of the summary is kept fixed but more entities are added, thus achieving a denser summary.

It is not clear to me what these entities are supposed to be. The authors claim to have computed them using spaCy, so I'm asking here if anyone can tell me what could be the relevant methods. Figure 2 contains, in green, a few examples of these entities: mostly, they are proper nouns such as Lotus, McLaren, Fernando Alonso, etc, but there are also a few specific details like rear-end damage, or five-second penalty.