GloVe learns embeddings from global co-occurrence statistics. This page lets students test real neighborhoods, compare similar words, and try analogy arithmetic such as king - man + woman.
GloVe does not predict one context word at a time like Skip-gram. It uses a matrix of global co-occurrence counts and learns vectors whose dot products explain those counts.
Type a word and fetch its nearest neighbors by cosine similarity. Good synonyms or related words should sit near the query word.
Build a new vector by adding positive terms and subtracting negative ones. The nearest neighbors of the result often reveal semantic structure.
This is the simplest way to see whether two words behave like synonyms or near-synonyms in the embedding space.