Main Concept
BERT is a Transformer-based model that reads text in both directions (left-to-right and right-to-left) simultaneously. Unlike GPT which generates text, BERT is designed for understanding and analyzing text.
Key Differences from GPT
| BERT | GPT | |
|---|---|---|
| Direction | Bidirectional | Unidirectional (left-to-right) |
| Purpose | Understanding text | Generating text |
| Use case | Classification, NER, similarity | Translation, summarization, code generation |
How It Works
BERT reads an entire sentence at once and considers context from both directions. This makes it better at understanding relationships between words, regardless of position.
Example: In the sentence “The bank provides loans,” BERT understands that “bank” means a financial institution because it sees the full context on both sides.
Common Uses
- Text classification (spam detection, sentiment analysis)
- Named Entity Recognition (NER)
- Question answering
- Similarity matching
AIF-C01 Context
Know that BERT and GPT are both Transformer-based, but serve different purposes: BERT understands, GPT generates. The exam may ask which one to use for a given task.