Generative AI

Question 1

Generative AI

Answer 1

AI that can produce high-quality content, including text, images, video, sound, synthetic data, etc.

Subset of Deep Learning, along with LLMs.

Question 1

How LLMs Work

Answer 1

Use Supervised Learning to repeatedly predict the next word in the sequence.

Question 2

Large Language Model

Key Tasks an LLM Can Carry Out

Answer 2

1. Reading
2. Writing
3. Chatting

Question 3

Common Use Cases for an LLM

Answer 3

• Brainstorming
• Drafting
• Proofreading
• Summarizing
• Analysis
• Sentiment
• Specialized Chatbots

Question 4

How to Deploy Chatbots

Answer 4

1. Deploy internally first
2. Deploy externally with human-in-the-loop
3. After deemed safe, allow chatbot to communicate directly with customers

May not be practical to have humans check every message.

Question 5

Hallucination

Hallucination

Answer 5

When the LLM generates new content that is not accurate, factual, or relevant.

Question 6

Causes of Hallucination

Answer 6

Causes:

• Ambiguous Input: if the user input is vague or unclear, the LLM might struggle to generate a relevant response, causing it to create content.
• Training Data: insufficient, biased, or otherwise flawed training data.
• Limitations of the model.

Question 7

Context Window

Answer 7

Amount of input data or tokens (words, characters, or other elements) that a model can process or “remember” at once. Determines how much of the preceding information the model can use to understand the current input. If the context window is too small, the model might miss long-range dependencies or context that could be crucial for accurate predictions, especially in tasks like language translation or summarization.

Question 8

Structured vs. Unstructured Data

Answer 8

Generative AI works better with unstructured versus structured (e.g., tabular) data.

Question 9

GPT Stands For

Answer 9

“Generative Pre-trained Transformer”

• Generative refers to generating new data in the form of text, images, etc.
• Pre-training is where the model learns from training data to generate probability distributions, usually with a combination of supervised and unsupervised ML techniques.
• Transformers are machine learning models built with neural networks that carry out the pre-training in a computationally effective manner.

Question 10

Sentiment Analysis

Answer 10

Process that uses natural language processing and machine learning techniques to identify opinions expressed in text input. Determines whether the attitude towards a particular topic, product, or service is positive, negative, or neutral.