Generative AI and Ethical Concerns / Prompt Engineering

Uses last 2 modules of the ppts from the AI class for short response questions

What major development in 2015 occurred with AI?

A machine learning algorithm learned to put labels into natural language descriptions

For an image/text generator to be able to respond so many different prompts, it needs

a massive diverse training set

If generative AI is used to produce content that directly copies or mimics copyrighted material, what happens?

it may constitute copyright infringement

Most copyright laws worldwide require a work to be created by a ____ to be eligible for protection

human author

If there’s significant human creativity involved in using the AI, art can be eligible for___

copyright protection

When AI is eligible for copyright protection, what is usually involved?

selecting specific prompts and settings, and adding human-created elements

There (are/aren’t) any definitive legal rulings yet on AI training data and copyright.

aren't

What are some ways people can distinguish AI-generated pictures versus real ones?

Reverse Image Search engines, Metadata analysis, Error level analysis, AI Image detection tools, etc

AI models are ____ based on the training data

biased

Microsoft created a Twitter AI bot named Tay, which was shut down within 24 hours. Why was this the case?

Users on twitter manipulated it into making offensive and inappropriate comments

In 2014, Amazon built an ML-based system to review job applicants' resumes with the aim of mechanizing the search for top talent. Why did this not work out in 2015?

The system was found to be biased against women, as it learned from resumes submitted to the company over a ten-year period, which were predominantly male.

In 2019, Apple Card was creating credit limit decisions for its users. Why did this not work out?

The algorithm lacked transparency, which made it difficult for users to understand its decisions.

In 2018, Uber self-driving car struck and killed a woman named Elaine Herzberg as she walked her bicycle across a road in Tempe, Arizona. How did this lead to ethical concerns?

The incident raised questions about accountability of the AI system.

Conversational AI’s collect what kind of information?

Personal Data, Health data, Financial data, and personal preference

In 2017, Saudi Arabia granted citizenship to Sophia, a humanoid social robot, making it the first robot in the world to receive such recognition. Why did this spark discussion?

This sparked discussion about the legal and ethical implications of granting citizenship to non-human entities and the potential impact on human rights.

How much energy did GPT-3’s training consume?

175 billion parameters

The computations involved in adjusting the model's parameters require…

sustained high-performance computing power and significant energy consumption.

What is the environmental impact of AI research doing?

it significantly contributes to carbon emissions and resource depletion

AI can replace which kinds of jobs?

those involving routine, repetitive, and rule based activities.

How can AI improve sectors of jobs like manufacturing, data entry, and customer service?

AI can improve efficiency, reduce human error, and lower costs.

What types of jobs is AI less likely to replace?

jobs that require creativity, critical thinking, emotional intelligence, and complex decision-making.

What makes music a human art form?

It reflects emotions, experiences, and stories in a way that AI currently struggles to replicate

How can AI be used with music moving forward?

they can be a tool that augments musicians rather than replaces them.

Do AI’s have a reasoning mechanism?

AI systems utilize algorithms and data patterns to mimic reasoning, but they lack true understanding and consciousness.

What is NLP (Natural Language processing)?

a branch of artificial intelligence (AI) that focuses on enabling computers to understand, interpret, and generate human language, both written and spoken

What are some attributes of NLP?

they process large amounts of language data, enabling machines to read or listen and respond intelligently.

Why is NLP Important?

It enables seamless interaction between humans and computers, making technology more accessible and efficient

What are some examples of everyday impact of NLP?

Healthcare, Fraud detection, Personalized reccomendations

What are the 2 key steps in NLP?

Understanding Language (NLU) and Generating Responses (NLP)

What is natural language understanding?

computers breaking down human language into structures they can process

What is natural language generation?

generating a response that makes sense after breaking down language

What are 4 aspects of Natural Language Understanding?

Tokenization, Part of Speech Tagging, Parsing, and Named Entity Recognition

What are 3 aspects of Natural Language Generation?

Text summarization and translation, and dialogue generation

What are the core components of NLP?

Syntax, Semantics, and Contextual Understanding

What is Syntax?

Analyzing how words are arranged in a sentence to understand the intended meaning.

What is Semantics/contextual understanding?

Understanding the meaning of words and sentences in context.

What are Rule-Based Systems? (Early NLP)

predefined set of rules/linguistic patterns

if-then rules are applied to text to detect patterns/keywords

What are machine learning models?

training on large datasets to recognize patterns in text

What are some examples of Machine Learning Models?

Naive Bayes - used for text classification
Support Vector Machines - Used for sentiment analysis

What are deep learning models? (Neural Networks)

mimic the way the human brain processes info and learns complex patterns from data

What are the benefits of deep learning models?

they handle massive amounts of unstructured text data and capture context and meaning over longer passages of text

What are the downsides of deep learning models?

require vast amts of data/computing power and are harder to interpret the decision making process

What are transformers and pre-trained models? (Modern NLP)

They use self attention mechanisms to process all words in a sentence simultaneously, which helps for understanding very long text

What are some examples of modern NLP?

BERT - can be fine tuned for tasks like question answering, text classification, and NER
GPT - excels in text generation, and generates text based on a given prompt

What are the benefits of Transformers and pre-trained models?

superior performance on a wide range of NLP tasks and can be fine-tuned for specific tasks with less data (no extensive retraining)

What are the downsides of Transformers and pre-trained models?

they produce biased outputs and require significant computational resources

What are the challenges in NLP?

Ambiguity - words having multiple meanings
Context - understanding the broader conversation
Cultural Nuances - languages use idioms, slang, etc

What is a prompt?

A piece of text or a set of instructions given to a Large Language Model (LLM) to trigger a desired response

What is prompt engineering?

The process of crafting and refining prompts to guide the model’s output and optimize accuracy and relevance

What are the benefits of prompt engineering?

improves accuracy, facilitate complex tasks, identifies model constraints, and improves output quality

What is finetuning?

a technique in which the parameters of a pre-trained model are modified using new data

What are the 6 main components of a prompt?

Instruction, Context, Input, Output format, Example, Clue

What are cues in ai output?

when an AI doesn’t answer your question, but just appends text to it.

What are some good practice in prompt engineering?

Clear and concise language - Clarity
Establishing context/including relevant info - Context
Being specific/using examples - Precision
Assign a persona - Persona/Roles

What is a Zero-Shot?

asking the model to solve a task without any specific preparation for it.

What is a few-shot?

Providing a model with a few examples of a task to guide its responses.

What is the Chain-Of-Thought technique in AI?

A method where the model is prompted to reason through a problem step-by-step, enhancing its ability to produce accurate responses.

What is the Self Consistency technique in AI?

A technique that involves generating multiple responses to a given prompt and selecting the most consistent or frequent answer, improving result reliability.

What is the Tree of Thought technique in AI?

A method that organizes reasoning processes in a branching structure, using search algorithms and allowing models to explore various pathways and solutions to a problem.

What is the prompt chaining technique in AI?

A method that links multiple prompts together, guiding the model through a sequence of tasks to improve context understanding and response accuracy.