2. WHAT IS NLP?
• Natural Language Processing (NLP) is a field of artificial intelligence that focuses on
the interaction between computers and human languages. It involves the
development of algorithms and models to enable computers to understand,
interpret, and generate human language. NLP has applications in various areas such
as machine translation, sentiment analysis, text summarization, and speech
recognition.
• Chat GPT, Sora AI, Siri, and Google Translate are some examples of Natural
Language Processing.
2
Fig.1
3. Things NLP Can Do:
• Sentiment Analysis
• Machine Translation
• Question Answering
• Text Summarization
• Language Generation
NLP Challenges:
• Multilinguality
• Multimodality
• Handling Noisy Data
• Ambiguity
• Understanding Context
3
4. HOW DOES NLP WORK IN CHAT GPT?
1.Tokenization:
Algorithm/Tool: WordPiece Tokenization or Byte Pair Encoding
Description: Splits the input text into smaller units called tokens, often
based on subword units or characters.
2.Word Embeddings:
Algorithm/Tool: Word2Vec, GloVe, or FastText
Description: Converts tokens into dense vector representations capturing
semantic relationships between words.
4
Fig.2
5. 3.Transformer Architecture:
Algorithm/Tool: Transformer architecture (e.g., OpenAI GPT, BERT)
Description: Using neural networks to find dependencies between words and
contextual information.
4.Training:
Algorithm/Tool: Backpropagation, Stochastic Gradient Descent (SGD).
Description: The model is trained with a large dataset of human conversations or
text data to help it understand the language and context. Similar to what we do to
babies.
5.Generation:
Algorithm/Tool: Beam Search, Top-k Sampling, Top-p Sampling.
Description: Generates a response by predicting the most likely next word based
on learned patterns and context, to assist the user in communicating.
5
6. 6.Decoding:
Algorithm/Tool: Greedy Decoding, Beam Search, Sampling techniques
Description: Converts the sequence of predicted tokens into human-readable
text, selecting the most likely sequence or sampling diverse sequences based on
different decoding strategies.
7.Fine-Tuning:
Algorithm/Tool: Transfer Learning, Fine-Tuning
Description: Retrains the model on a smaller dataset relevant to the specific
task or domain, updating parameters to improve performance in particular
contexts.
6
7. BIBLIOGRAPHY
• Intro To NLP: An informative PowerPoint presentation by Sergei Nirenberg.
https://www.seas.upenn.edu/~eeaton/teaching/cmsc471_fall07/slides/IntroToNLP.ppt
• Chat GPT: Tools used in NLP for ChatGPT, and SoraAI.
https://chat.openai.com/
• Images: https://tezo.com/
https://commons.wikimedia.org/
https://www.vecteezy.com/