The WWW of Gen AI - AI Summit 2024

DevLearn 24

the www of gen ai

AI SUMMIT

Maikel Right

Associate Director of Insturctional Learning Technologies FIU Online Faculty Fellow Honors College & Liberal Studies

• Unashamed Millienial
• Foster Parent to Teens...I mean "Young Adults"
• Digital Consultant
• Academic Intergrity/Student Conduct Commitee

+ CONNECT

how we're going to do this

01. What is gen ai? & what it isn't 02. Begin at the end 03. privacy conCerns 04. bias 05. Responsible use 06. what do we do now?

a category of AI that includes various systems capable of producing content in response to prompts.

Potter, R. L. (2024)

There were 7 months and 23 days between the launches of ChatGPT-3.5 and ChatGPT-4

"ChatGPT performed at or near the passing threshold for all three exams without any specialized training or reinforcement," they wrote, noting that the tool was able to demonstrate "a high level of concordance and insight in its explanations.""These results suggest that large language models may have the potential to assist with medical education, and potentially, clinical decision-making,"

Michael DePeau-Wilson (2023)

Creative

invisible

Everywhere

Generated more, cheaper, and better ideas than students

ChatGPT has passed the Turing test

Adopted faster that PCs and the internet

accessible

complete

not google

The BEST models are those that are paid

The current version you use today is the worst version there will ever be

Not a serach engine............yet

most questions begin with the end in mind

"Let's start with the apocalypse"

1. Answer Questions on Microsoft Forms
2. Complete our collective Doomsday Mad Lib
3. Ask ChatGPT to write our story
4. Share myths, misconceptions and truths

discussion through GRoupstorytelling

Imagine AI collects _______ data from students, like ________, and shares it with _______ without informing them.

Privacy concerns

Where we are when it comes to privacy concerns

1. Current litigation with Google"taken all our personal and professional information, our creative and copywritten works, our photographs, and even our emails—virtually the entirety of our digital footprint"
2. As companies build, train, and deploy generative AI solutions, they increase their liability and exposure to class action claims related to data privacy
3. While there are numerous lawsuits related to generative AI, most focus on copyright infringement rather than privacy issues.

let's talk about weights in ai

• ChatGPT doesn't store exact phrases to copy later. Instead, it learns general patterns of language during training.

• The weights represent what the AI has learned about these patterns.

• Weights allow ChatGPT to combine information in new ways, rather than just repeating memorized text.

This is why it can generate original responses to questions it hasn't seen before.

let's talk about weights in ai

• Weights are actually numbers that strengthen or weaken connections in the AI's neural network. They're not words or phrases, but mathematical representations of learned patterns.

• When you ask ChatGPT a question, it uses its weights to generate a response word by word, not by looking up pre-written answers

Tell me how closely this sounds to how skilled workers refine their craft!?

So what do we do about privacy?

• Outline the boundaries of "publicly available" information and whether companies need explicit consent to use such data for AI training.

• Be transparent
• Make our own GPTs???
• Continuously monitor AI use and update policies as the technology and privacy landscape evolve.

"Imagine an AI program that recommends students for advanced courses based on their _________, but it tends to favor students from _________ communities, while overlooking qualified students from _________ backgrounds."

Bias

Where we are when it comes to Bias

1. Open-source LLMs like Llama 2 tended to assign more diverse, high-status jobs to men (e.g., engineer, teacher, doctor), while frequently relegating women to traditionally undervalued or socially-stigmatized roles (e.g., domestic servant, cook).
2. When prompted to complete sentences beginning with "a gay person is...", 70% of content generated by Llama 2 and 60% by GPT-2 was negative.
3. All tested models (GPT-4, GPT-4-turbo, GPT-3.5-turbo, GPT-3) exhibited cultural values resembling English-speaking and Protestant European countries.

48%

of HR managers admit bias affects their candidate choice.

Bias has always been an issue in many fields not excluding education. Research shows that educators display the same level of racial bias as the general population.

What do we do?

Keep humans in the loop

Experiment cautiously

Be intentional about Gen AI use

Just because Gen AI has limitations does not mean it will always nor that it cannot be helpful.

In particular for tasks that must meet legal requirements or involve potential biases.

Consider spaces and places where bias is normally an issue and reconsider relying solely on Gen AI there.

Imagine an AI tool used to predict _________ in students, but it falsely flags students with ______, leading to _______.

Responsible use

1. Recognize AI's biases and limitations.
1. Maintain human oversight in critical decision-making processes.
2. Use AI to augment, not replace, human intelligence.
3. Establish clear guidelines for AI use in different contexts.
2. Model and encourage low-risk AI experimentation.
1. Educate users about AI's potential errors and limitations.
2. Extensively use AI to understand its capabilities fully.
3. Communicate with AI as if it were human, but define its role clearly.

what does respons-ible use look like

Florida International University will achieve exceptional student-centered learning and upward economic mobility, produce meaningful research and creative activities, and lead transformative innovations locally and globally, resulting in recognition as a Top-50 public university.

FIU Mission Statement

Some of my favorite resources

• Perplexity AI
• Co-Intelligence by Ethan Molick
• Teaching with AI (YouTube)
• Planet Money/Indicator podcast
• Pi.ai
• Midjourney

Thank you!