BERT
In plain terms
A transformer model that reads text bidirectionally for understanding.
Going deeper
BERT learns deep language representations by considering context from both directions, excelling at understanding tasks like classification and search. It reads whole context at once.
Under the hood
Encoder-only and trained with masked-language modeling; great for understanding, not generation.
Examples
Search ranking, classification, NER, sentiment analysis, Google Search.
Think of it like...
Like a reader who considers the whole sentence at once, not just left to right.