What is One Hot Encoding?

One Hot Encoding is a vectorization technique where each word is represented by a binary vector. Only one position contains 1. All other positions contain 0.

Why One Hot Encoding?

Machines cannot understand words. Therefore we assign a unique vector to every word.

Step 1 : Create Vocabulary

Step 2 : Assign Positions

Step 3 : Generate Binary Vectors

Visualization

Large Example

Advantages

• Simple to Understand
• Easy to Implement
• Works Well for Small Vocabulary
• Good for Learning Fundamentals

Limitations

• Huge Memory Usage
• Sparse Matrix Problem
• No Semantic Meaning
• Cannot Understand Context
• Cannot Understand Similar Words

Sparse Matrix Problem

Imagine a vocabulary of 10000 words.

Most values are zero. This wastes memory.

Quick Quiz

1. What is One Hot Encoding?
2. Why do we use it?
3. What is a binary vector?
4. What is a sparse matrix?
5. What are its limitations?

Interview Questions

1. Explain One Hot Encoding.
2. How does One Hot Encoding work?
3. What is the Sparse Matrix Problem?
4. Why is One Hot Encoding inefficient?
5. Difference between One Hot Encoding and Bag of Words?

What is Bag of Words?

Bag of Words (BoW) is one of the most popular text vectorization techniques. Instead of representing a word as a unique binary vector like One Hot Encoding, BoW counts how many times each word appears in a document.

Why Do We Need Bag of Words?

One Hot Encoding only tells whether a word exists. It does not tell how important that word is.

Step 1 : Create Vocabulary

Step 2 : Count Word Occurrences

Step 3 : Create Vector

The count of each vocabulary word becomes the vector.

BoW Visualization

BoW for Multiple Documents

Vocabulary

Document-Term Matrix

Each row represents a document. Each column represents a vocabulary word.

Real World Example

Major Limitation of BoW

Bag of Words ignores word order.

Advantages of Bag of Words

• Simple to understand
• Easy to implement
• Works well with traditional ML algorithms
• Captures word frequency
• Fast processing

Disadvantages of Bag of Words

• Ignores word order
• Cannot understand meaning
• Cannot understand synonyms
• Sparse matrix problem
• Large vocabulary increases memory usage

One Hot Encoding vs Bag of Words

Python Example

Quick Quiz

1. What is Bag of Words?
2. Why is BoW better than One Hot Encoding?
3. What is a Document-Term Matrix?
4. How does BoW represent text?
5. What is the major limitation of BoW?

Interview Questions

1. Explain Bag of Words with an example.
2. How is BoW different from One Hot Encoding?
3. What is a Document-Term Matrix?
4. Why does BoW ignore context?
5. What are the limitations of Bag of Words?
6. How does CountVectorizer implement BoW?
7. What is the Sparse Matrix problem?
8. Can BoW understand semantics?

What is Count Vectorizer?

Count Vectorizer is a feature extraction technique provided by Scikit-Learn. It automatically converts text documents into a numerical matrix by counting the frequency of words.

Why Do We Need Count Vectorizer?

Creating a Bag of Words manually becomes difficult when dealing with thousands of documents and vocabulary terms. Count Vectorizer automatically:

• Creates Vocabulary
• Assigns Index Values
• Counts Word Frequencies
• Generates Feature Matrix
• Prepares Data for Machine Learning

Count Vectorizer Workflow

Example Dataset

Step 1 : Create Vocabulary

Count Vectorizer scans all documents and collects unique words.

Step 2 : Create Document-Term Matrix

Count Vectorizer counts occurrences of each word in every document.

Understanding the Feature Matrix

Each row represents a document. Each column represents a vocabulary word. The values indicate the frequency of occurrence.

How Count Vectorizer Works Internally

Important Parameters

Python Example

Expected Output

Advantages of Count Vectorizer

• Simple to use
• Automatically builds vocabulary
• Works well for small and medium datasets
• Fast implementation
• Useful for text classification tasks

Limitations of Count Vectorizer

• Ignores word meaning
• Cannot understand context
• Sparse matrix problem
• Large memory consumption
• Treats similar words as unrelated

Bag of Words vs Count Vectorizer

Real World Applications

• Email Spam Detection
• News Classification
• Sentiment Analysis
• Document Categorization
• Review Classification

Quick Quiz

1. What is Count Vectorizer?
2. How is it related to Bag of Words?
3. What is a Feature Matrix?
4. Why do we need Count Vectorizer?
5. What are its limitations?

Interview Questions

1. Explain Count Vectorizer.
2. How does Count Vectorizer work internally?
3. Difference between BoW and Count Vectorizer?
4. What is a Document-Term Matrix?
5. What are important Count Vectorizer parameters?
6. What is the sparse matrix problem?
7. Why is Count Vectorizer widely used?
8. What are the limitations of Count Vectorizer?

What are N-Grams?

N-Grams are contiguous sequences of N words extracted from a sentence. Unlike Bag of Words, which treats words independently, N-Grams preserve word order and capture local context.

Why Do We Need N-Grams?

Bag of Words ignores word order. Sometimes word order completely changes the meaning.

Types of N-Grams

1. Unigram (N = 1)

Each individual word is treated as a token.

Generated Unigrams

2. Bigram (N = 2)

Two consecutive words are grouped together.

Generated Bigrams

3. Trigram (N = 3)

Three consecutive words are grouped together.

Generated Trigram

Large Example

Unigrams

Bigrams

Trigrams

N-Gram Generation Visualization

Why N-Grams are Important in Sentiment Analysis?

Consider the sentence:

Bag of Words View

The model may incorrectly interpret "good" as positive.

Bigram View

The phrase meaning is preserved.

Bigram Feature Matrix

Creating N-Grams using Count Vectorizer

(2,2) → Only Bigrams

(1,2) → Unigrams + Bigrams

Understanding ngram_range

Advantages of N-Grams

• Preserves Word Order
• Captures Local Context
• Improves Sentiment Analysis
• Better than Pure Bag of Words
• Works Well with Text Classification

Disadvantages of N-Grams

• Vocabulary grows rapidly
• Consumes more memory
• Sparse matrix problem
• Still cannot understand semantics
• Computationally expensive for large datasets

Bag of Words vs N-Grams

Applications of N-Grams

• Sentiment Analysis
• Machine Translation
• Spell Checking
• Text Prediction
• Language Modeling
• Speech Recognition
• Search Engines

Quick Quiz

1. What is an N-Gram?
2. What is a Bigram?
3. What is a Trigram?
4. Why are N-Grams useful?
5. What problem of Bag of Words do N-Grams solve?
6. What does ngram_range=(1,2) mean?
7. Why are N-Grams useful in sentiment analysis?

Interview Questions

1. What are N-Grams?
2. Explain Unigram, Bigram and Trigram.
3. How do N-Grams preserve context?
4. What is the limitation of Bag of Words?
5. How does CountVectorizer generate N-Grams?
6. What is ngram_range?
7. Why do N-Grams increase vocabulary size?
8. What are the applications of N-Grams?
9. Can N-Grams understand semantic meaning?
10. Why are N-Grams useful in sentiment analysis?

What is TF-IDF?

TF-IDF is one of the most important feature extraction techniques in Natural Language Processing. While Bag of Words and Count Vectorizer only count words, TF-IDF identifies which words are important within a document.

Why Do We Need TF-IDF?

Suppose we have thousands of documents. Words like:

appear in almost every document. These words do not help distinguish documents.

Real Life Example

The word "is" appears in all documents. The word "Amazing" appears in only one document.

Term Frequency (TF)

Term Frequency measures how frequently a word appears in a document.

Example

Total Words = 3 Frequency of NLP = 2

Document Frequency (DF)

Document Frequency represents the number of documents containing a particular word.

Inverse Document Frequency (IDF)

IDF measures how rare a word is across all documents.

Example

Total Documents = 3 DF(Amazing) = 1

Final TF-IDF Formula

A word receives a high score when:

• It appears frequently in a document.
• It appears rarely across all documents.

Complete Worked Example

Step 1 : Calculate TF

Step 2 : Calculate IDF

Step 3 : Calculate TF-IDF

TF-IDF Workflow

TF-IDF Matrix

Python Example

Advantages of TF-IDF

• Identifies important words.
• Reduces common word influence.
• Improves text classification.
• Works well with Machine Learning.
• Simple and efficient.

Limitations of TF-IDF

• Cannot understand word meaning.
• Cannot understand context.
• Cannot understand synonyms.
• Ignores sentence structure.
• Still produces sparse matrices.

Comparison of Traditional Vectorization Techniques

Traditional NLP Vectorization Journey

Quick Quiz

1. What does TF-IDF stand for?
2. What is Term Frequency?
3. What is Document Frequency?
4. What is Inverse Document Frequency?
5. Why is TF-IDF better than Bag of Words?
6. Write the TF-IDF formula.
7. Why do common words get lower scores?
8. What are the limitations of TF-IDF?

Interview Questions

1. Explain TF-IDF with an example.
2. What is the difference between TF and IDF?
3. Why is TF-IDF important in NLP?
4. How does TF-IDF improve text classification?
5. What are the limitations of TF-IDF?
6. Can TF-IDF understand semantics?
7. What is the difference between CountVectorizer and TfidfVectorizer?
8. Why does TF-IDF reduce common word importance?
9. What is sparse matrix representation?
10. When should TF-IDF be preferred?

Why Do We Need Word Embeddings?

Traditional Vectorization techniques such as:

• One Hot Encoding
• Bag of Words
• Count Vectorizer
• N-Grams
• TF-IDF

focus mainly on counting words. They do not understand:

• Meaning
• Context
• Relationships
• Synonyms
• Semantic Similarity

Problem with Traditional Vectorization

The model knows these are different words. But it does not know their relationships.

What is Semantic Understanding?

Semantic Understanding means understanding the meaning of words and their relationships.

Humans naturally understand these relationships. Word Embeddings help machines learn them.

What is a Word Embedding?

A Word Embedding is a dense numerical vector that represents the meaning of a word.

Sparse Vectors vs Dense Vectors

Word Embedding Visualization

Words with similar meanings appear close together in vector space.

How Word Embeddings Work

The model observes which words frequently appear together.

Context Learning Example

The model notices:

• King and Queen appear in similar contexts
• Kingdom appears near both words
• Prince appears in related contexts

Famous Word Embedding Analogy

Word Embeddings can learn relationships mathematically.

Types of Word Embeddings

Introduction to Word2Vec

Word2Vec is one of the most popular embedding algorithms. It learns word meanings based on surrounding context.

Applications of Word Embeddings

• ChatGPT
• Machine Translation
• Question Answering
• Search Engines
• Recommendation Systems
• Sentiment Analysis
• Speech Recognition
• Text Summarization

Traditional NLP vs Modern NLP

Quick Quiz

1. What is a Word Embedding?
2. Why are embeddings better than TF-IDF?
3. What is semantic understanding?
4. What is a dense vector?
5. Why are King and Queen close in embedding space?
6. Name three embedding techniques.
7. What is context learning?

Interview Questions

1. What are Word Embeddings?
2. Why do we need embeddings?
3. Difference between TF-IDF and Word Embeddings?
4. What are dense vectors?
5. What is semantic similarity?
6. Explain the King-Man+Woman=Queen analogy.
7. What are the types of embeddings?
8. How do embeddings capture meaning?
9. Why are embeddings important for LLMs?
10. What is embedding space?

What is Word2Vec?

Word2Vec is a Deep Learning based Word Embedding algorithm introduced by Google in 2013. It converts words into dense vectors while preserving semantic meaning.

Why Was Word2Vec Introduced?

Traditional techniques such as:

• One Hot Encoding
• Bag of Words
• Count Vectorizer
• TF-IDF

cannot understand relationships between words.

Main Idea Behind Word2Vec

Words that appear in similar contexts usually have similar meanings.

Since King and Queen appear in similar surroundings, Word2Vec learns that they are related.

How Word2Vec Works

Understanding Context

Word2Vec learns from neighboring words.

Word2Vec Architectures

CBOW (Continuous Bag of Words)

CBOW predicts the target word using surrounding context words.

Using: I, Love, Processing Predict: Natural

Skip-Gram

Skip-Gram does the opposite. It predicts surrounding words using the target word.

Predict: Love Language

CBOW vs Skip-Gram

Window Size

Window Size determines how many neighboring words are considered as context.

Window Size = 2

Generated Word Vectors

Notice how both vectors are very similar.

Semantic Relationships Learned by Word2Vec

Word2Vec Training Process

Python Example using Gensim

Advantages of Word2Vec

• Captures semantic meaning
• Dense vectors
• Memory efficient
• Understands word relationships
• Works well with NLP applications

Limitations of Word2Vec

• Same word always gets same vector
• Cannot handle multiple meanings effectively
• Needs large corpus
• Context understanding is limited

Applications of Word2Vec

• Search Engines
• Recommendation Systems
• Chatbots
• Machine Translation
• Sentiment Analysis
• Question Answering Systems

Quick Quiz

1. What is Word2Vec?
2. Why was Word2Vec introduced?
3. What is CBOW?
4. What is Skip-Gram?
5. What is Window Size?
6. Why are King and Queen close in vector space?
7. What are the advantages of Word2Vec?

Interview Questions

1. Explain Word2Vec.
2. How does Word2Vec learn word meanings?
3. Difference between CBOW and Skip-Gram?
4. What is Window Size?
5. Why are embeddings dense vectors?
6. Explain the King − Man + Woman = Queen analogy.
7. What are the limitations of Word2Vec?
8. Why is Word2Vec better than TF-IDF?
9. What is semantic similarity?
10. How is Word2Vec trained?

What is CBOW?

CBOW (Continuous Bag of Words) is one of the two architectures of Word2Vec. Its goal is simple:

CBOW learns word meanings by observing neighboring words.

Real Life Intuition

Humans can often guess a missing word using surrounding words.

Most people will predict:

CBOW trains a neural network to perform exactly this task.

Main Idea Behind CBOW

Example Sentence

Assume:

Understanding Window Size

Window Size determines how many neighboring words are used as context.

Window Size = 1

Window Size = 2

CBOW Architecture

Input Layer

The Input Layer receives context words.

Context words are converted into vectors.

Hidden Layer

The Hidden Layer learns relationships between words.

This layer gradually learns:

• Semantic Meaning
• Word Relationships
• Context Information
• Word Similarities

Output Layer

The Output Layer predicts the most probable target word.

Complete CBOW Workflow

Training Example

Mathematical View

CBOW attempts to maximize:

Meaning: Probability of Target Word given the surrounding Context Words.

CBOW Visualization

The model learns that these context words frequently surround "Natural".

Why CBOW Works

Words appearing in similar contexts often have similar meanings.

Since King and Queen appear in similar surroundings, their embeddings become similar.

Advantages of CBOW

• Fast Training
• Efficient on Large Datasets
• Good General Embeddings
• Less Computational Cost
• Works Well for Frequent Words

Limitations of CBOW

• Poor for Rare Words
• Uses Fixed Context Window
• Cannot Handle Multiple Meanings Well
• Context Understanding is Limited

CBOW vs Skip-Gram

Python Example using Gensim

Important Note

Applications of CBOW

• Search Engines
• Recommendation Systems
• Document Classification
• Chatbots
• Machine Translation
• Question Answering Systems

Quick Quiz

1. What is CBOW?
2. What does CBOW predict?
3. What is Window Size?
4. What is the role of the Hidden Layer?
5. What does sg=0 mean?
6. Why is CBOW faster than Skip-Gram?
7. What are the limitations of CBOW?

Interview Questions

1. Explain CBOW architecture.
2. How does CBOW work?
3. What is the difference between CBOW and Skip-Gram?
4. What is Window Size?
5. Why is CBOW faster?
6. How are embeddings learned in CBOW?
7. What does sg=0 represent?
8. What are the advantages of CBOW?
9. What are the limitations of CBOW?
10. When should CBOW be preferred?

What is Skip-Gram?

Skip-Gram is one of the two architectures of Word2Vec. Unlike CBOW, which predicts a target word from context words, Skip-Gram predicts surrounding context words using the target word.

Real Life Intuition

Suppose we know the word:

We can often predict nearby words such as:

Skip-Gram trains a neural network to perform this prediction.

Main Idea Behind Skip-Gram

Example Sentence

Assume:

Understanding Window Size

Window Size determines how many neighboring words are predicted.

Window Size = 1

Window Size = 2

Skip-Gram Architecture

Input Layer

The Input Layer receives the target word.

The target word is converted into a numerical vector.

Hidden Layer

The hidden layer learns semantic relationships among words.

• Word Meaning
• Word Similarity
• Context Information
• Relationships

Output Layer

The Output Layer predicts nearby context words.

Training Example

Complete Skip-Gram Workflow

Mathematical View

Skip-Gram tries to maximize:

Meaning: Probability of observing surrounding words given a target word.

How Skip-Gram Learns Meaning

The model notices:

• King appears near rules and kingdom
• Queen appears near rules and kingdom

Why Skip-Gram is Better for Rare Words?

Each target word generates multiple training examples.

This helps the model learn rare words more effectively.

CBOW vs Skip-Gram

CBOW vs Skip-Gram Visualization

Python Example using Gensim

Important Note

Advantages of Skip-Gram

• Learns better embeddings
• Handles rare words effectively
• Captures semantic relationships
• Produces high-quality vectors
• Suitable for large corpora

Limitations of Skip-Gram

• Slower training
• Requires more computation
• Needs large datasets
• Cannot fully understand sentence meaning

Applications of Skip-Gram

• Search Engines
• Recommendation Systems
• Machine Translation
• Chatbots
• Text Classification
• Question Answering Systems
• Semantic Search

Quick Quiz

1. What is Skip-Gram?
2. How is Skip-Gram different from CBOW?
3. What does Skip-Gram predict?
4. What is Window Size?
5. Why is Skip-Gram better for rare words?
6. What does sg=1 mean?
7. What are the advantages of Skip-Gram?

Interview Questions

1. Explain Skip-Gram architecture.
2. How does Skip-Gram work?
3. Difference between CBOW and Skip-Gram?
4. Why is Skip-Gram better for rare words?
5. What is Window Size?
6. What does sg=1 represent?
7. How are embeddings learned in Skip-Gram?
8. What are the advantages of Skip-Gram?
9. What are the limitations of Skip-Gram?
10. When should Skip-Gram be preferred?

What is GloVe?

GloVe (Global Vectors for Word Representation) is a word embedding technique developed by Stanford University. It creates meaningful word vectors by analyzing how frequently words appear together across an entire corpus.

Why Was GloVe Introduced?

Word2Vec is excellent at learning local context. However, it does not fully utilize the overall statistics of the corpus.

GloVe studies these global relationships across the entire dataset.

Main Idea Behind GloVe

Words that frequently occur together should have similar vector representations.

What is a Co-occurrence Matrix?

A Co-occurrence Matrix stores how often words appear together.

Local Context vs Global Context

Matrix Factorization

The Co-occurrence Matrix can become extremely large. GloVe compresses this information into smaller dense vectors.

GloVe Mathematical Formula

Where:

• wᵢ = Word Vector
• wⱼ = Context Vector
• bᵢ = Word Bias
• bⱼ = Context Bias
• Xᵢⱼ = Co-occurrence Count

Simple Explanation of the Formula

The formula tries to ensure:

The model continuously adjusts vectors until the mathematical relationship matches real-world word co-occurrences.

Example: Learning Relationships

Because these words appear in similar contexts, their vectors become similar.

Famous GloVe Analogy

These relationships emerge naturally from word co-occurrence statistics.

GloVe Training Process

Embedding Space Visualization

Words with similar meanings cluster together.

Word2Vec vs GloVe

Using Pretrained GloVe Embeddings

Advantages of GloVe

• Captures semantic meaning
• Uses global corpus statistics
• Produces high-quality embeddings
• Learns word relationships
• Efficient for large datasets
• Excellent for NLP applications

Limitations of GloVe

• Cannot handle unseen words
• Static embeddings
• One word = One vector
• Cannot handle multiple meanings effectively
• Requires large training corpora

Applications of GloVe

• Chatbots
• Machine Translation
• Search Engines
• Recommendation Systems
• Question Answering Systems
• Text Classification
• Sentiment Analysis
• Information Retrieval

GloVe Summary

Quick Quiz

1. What does GloVe stand for?
2. What is a Co-occurrence Matrix?
3. Why was GloVe introduced?
4. How is GloVe different from Word2Vec?
5. What is Matrix Factorization?
6. Why are King and Queen close in vector space?
7. What are the limitations of GloVe?

Interview Questions

1. Explain GloVe Embeddings.
2. How does GloVe work?
3. What is a Co-occurrence Matrix?
4. Difference between Word2Vec and GloVe?
5. What is Matrix Factorization?
6. Why does GloVe produce better semantic vectors?
7. What are the advantages of GloVe?
8. What are its limitations?
9. Explain the King − Man + Woman = Queen analogy.
10. When would you use GloVe instead of Word2Vec?

What is FastText?

FastText is a Word Embedding algorithm developed by Facebook AI Research (FAIR). It improves Word2Vec by learning not only complete words but also parts of words called subwords.

Why Was FastText Introduced?

Word2Vec and GloVe have a major limitation.

Traditional embeddings treat these as completely different words.

What is OOV (Out Of Vocabulary)?

OOV words are words that were never seen during training.

Word2Vec: ❌ Cannot generate embedding FastText: ✔ Can generate embedding

Main Idea Behind FastText

FastText breaks words into smaller character-level chunks called N-Grams.

Character N-Grams

Consider the word:

FastText generates character N-Grams.

< and > represent beginning and ending boundaries.

Word Construction Visualization

Example: Unknown Word Handling

Suppose the model has seen:

Now a new word appears:

FastText can still understand it because it recognizes:

FastText Architecture

Training Process

1. Collect Text Corpus
2. Tokenize Text
3. Generate Character N-Grams
4. Train Neural Network
5. Learn Subword Embeddings
6. Combine Subwords into Final Word Vector

Word2Vec vs GloVe vs FastText

Morphological Understanding

Morphology refers to the structure of words.

FastText understands that all these words are related because they share similar character patterns.

FastText Embedding Example

Notice that both vectors remain very similar.

Python Example

Real World Applications

• Chatbots
• Search Engines
• Spell Correction
• Machine Translation
• Text Classification
• Question Answering
• Recommendation Systems
• Multilingual NLP

Advantages of FastText

• Handles OOV Words
• Uses Character-Level Information
• Captures Word Structure
• Better for Rare Words
• Works Well on Small Datasets
• Excellent for Morphologically Rich Languages

Limitations of FastText

• Larger Model Size
• More Memory Usage
• Slower Training than Word2Vec
• Still Produces Static Embeddings
• Cannot Fully Understand Sentence Meaning

Evolution of Word Embeddings

Quick Quiz

1. What is FastText?
2. Why was FastText introduced?
3. What is OOV?
4. What are Character N-Grams?
5. How does FastText handle unknown words?
6. Difference between Word2Vec and FastText?
7. What is morphology?
8. What are the advantages of FastText?

Interview Questions

1. Explain FastText.
2. What is the OOV problem?
3. How does FastText solve OOV?
4. What are Character N-Grams?
5. Difference between Word2Vec, GloVe and FastText?
6. What is morphological understanding?
7. Why is FastText useful for rare words?
8. What are the limitations of FastText?
9. How does FastText generate embeddings?
10. When should FastText be preferred?

The Next Evolution in NLP

We have already studied:

These methods improved NLP significantly. However they still have a major limitation.

Problem with Traditional Embeddings

Consider the word:

Humans understand that "bank" has different meanings.

Polysemy Problem

Polysemy means:

Static vs Contextual Embeddings

What are Contextual Embeddings?

Contextual Embeddings generate different vectors for the same word depending on the sentence.

Introduction to Transformers

Transformers are Deep Learning architectures introduced by Google in 2017. They completely changed NLP.

Why Transformers Replaced RNNs and LSTMs?

Attention Mechanism

Attention helps the model focus on important words while processing a sentence.

When processing:

the model gives more attention to:

Self-Attention

Self-Attention allows every word to look at every other word in the sentence.

Self-Attention Example

Question: Who was tired?

Self-Attention helps identify such relationships.

Transformer Architecture

Encoder and Decoder

Transformer Family

BERT

BERT stands for:

BERT reads sentences in both directions.

GPT

GPT stands for:

GPT predicts the next word repeatedly to generate text.

Modern LLMs

Modern Large Language Models are built using Transformer architectures.

• GPT-4
• GPT-5
• Gemini
• Claude
• Llama
• Mistral

Evolution of NLP

Quick Quiz

1. What are Contextual Embeddings?
2. What is Polysemy?
3. Why do static embeddings fail?
4. What is a Transformer?
5. What is Attention?
6. What is Self-Attention?
7. Difference between BERT and GPT?
8. Why are Transformers important?

Interview Questions

1. Explain Contextual Embeddings.
2. Difference between static and contextual embeddings?
3. What problem do Transformers solve?
4. What is Self-Attention?
5. Why are Transformers faster than RNNs?
6. Explain Encoder and Decoder.
7. Difference between BERT and GPT?
8. What is Polysemy?
9. How do LLMs use Transformers?
10. Why did Transformers revolutionize NLP?