Introduction
Figure 1
NLP is an interdisciplinary field, and LLMs are
just a subset of it
Figure 2
ChatGPT Just Works! Does it…?
Figure 3
GPT-4 Tokenization Example
Figure 4
Unsupervised Learning
Figure 5
Supervised Learning
Figure 6
A taxonomy of NLP Tasks
Figure 7
Levels of Language
Figure 8
Diagram showing building blocks of
language
From words to vectors
Figure 1
After seeing thousands of examples, each layer
represents different “features” that maximize the success of the task,
but they are not human-readable. The last layer acts as a classifier and
outputs the most likely label given the input
Figure 2
Figure 3
Figure 4
Transformers: BERT and Beyond
Figure 1
Transformer Architecture
Figure 2
Figure 3
BERT Architecture
Figure 4
The Encoder Self-Attention Mechanism
Figure 5
BERT Language Modeling
Figure 6
BERT as an Emotion Classifier
Figure 7
BERT as an Emotion Classifier
Figure 8
An example for a classifier of Cats and Dogs.
Source: Wikipedia
Figure 9
Figure 10
BERT as an NER Classifier
Using large language models
Figure 1
Figure 2
Figure 3
Figure 4
Figure 5
Figure 6
Generative LLMs correspond to the Decoder
component of the Transformer architecture