Artificial Intelligence and Augmented Reality workshop for DePaul Coleman Entrepreneurship Center students.

You can download a PDF of the presentation handout here.

I used Google’s NotebookLM service to provide additional notes from the workshop transcript. These notes have been created by an LLM, so may contain inaccuracies (on top of any inaccuracies I may have created).

NotebookLM also creates an entertaining podcast recording, which you can listen to here.

Timeline of Events
This source does not describe a series of events occurring over time. Instead, it is a transcript of a lecture or presentation about AI, large language models, and augmented reality, delivered by a speaker named James. The presentation covers various concepts and technologies related to AI and their potential applications and implications. There is no chronological narrative or story presented in the transcript.

Cast of Characters
James: The speaker of the presentation. James appears to be an expert in AI, large language models, and augmented reality. He is likely a professor or lecturer at DePaul University, given his references to students, entrepreneurs, and DePaul usernames. He aims to educate the audience about different AI concepts, their current limitations, potential benefits, and ethical concerns. He offers advice on navigating the AI landscape, particularly regarding privacy, copyright, and career opportunities.

Other Participants: The transcript mentions an audience composed of students, some of whom identify as entrepreneurs. They engage with James by asking questions and offering observations, demonstrating their interest in learning about AI and its implications.

Mentioned Figures:

• Steve Jobs: Former CEO of Apple, known for his vision of "lickable" technology – devices so appealing that users would desire them like candy.
• Demis Hassabis: CEO and co-founder of DeepMind, acquired by Google. He is considered one of the most influential figures in the AI space.

Companies:

• Meta: The company behind Facebook, Instagram, and the Meta Ray-Ban smart glasses.
• Google: A technology giant heavily invested in AI, with products like Google Glass and Gemini.
• Microsoft: A major technology company developing AI products like Copilot and previously, Recall.
• OpenAI: The company behind the popular large language model, ChatGPT.
• Adobe: A software company known for its creative tools, integrating AI into its products.
• Amazon: A multinational technology company dabbling in AR with products like Echo Frames.
• Snap: The company behind Snapchat, developing AR glasses, Snapchat Spectacles.

This transcript provides insights into the current state of AI and its potential impact on various industries. It highlights the need for awareness about ethical considerations, privacy risks, and copyright issues surrounding AI development and deployment. The presentation also encourages students to explore the opportunities and challenges in this rapidly evolving field.

Source: Transcript of a presentation by James, an individual whose full name and credentials are not provided in the excerpt, delivered at DePaul University. The presentation focuses on providing students with an overview of Artificial Intelligence (AI), Large Language Models (LLMs), and Augmented Reality (AR).

Date: Unknown

Key Themes and Ideas:

• Defining AI and its Categories:
• James emphasizes the difficulty of defining AI due to its broad scope, noting that literature often describes it as the "ability for something to act in a way that mimics human intelligence."
• He introduces three categories of AI:
• Narrow/Weak AI: Completes a single task, representing the current state of most AI encountered today.
• Artificial General Intelligence (AGI): Possesses cognitive abilities like perceiving, thinking, and having an internal mental model, which is the current target of major AI companies.
• Artificial Superintelligence (ASI): Surpasses human intelligence in all abilities, often referred to as "the singularity," which is a hypothetical future state.
• Understanding LLMs:
• LLMs are described as systems that "respond or write like a human," not through understanding, but by calculating the statistical probability of the next word in a sequence.
• James stresses that LLMs "don't know the difference between things that are false and true," leading to potential problems with "hallucinations," which manifest as generating incorrect or fabricated information.
• He differentiates between "closed domain" and "open domain" hallucinations, the former being factual errors and the latter being outright fabrication.
• LLMs function by using vector databases to calculate the relationships between concepts (tokens) across hundreds of dimensions.
• James advises students to familiarise themselves with "foundation models" (base models available for use) and "frontier models" (cutting-edge models) to enhance their job prospects.
• Navigating Free and Paid LLMs:
• Free LLM options like Meta's LLaMa are "open source," allowing manipulation and customization. Platforms like Hugging Face host a vast number of free models.
• However, James cautions that free LLMs retain user data for training future models, posing risks for entrepreneurs who input sensitive business information.
• He highlights the data retention policies of various LLMs: Claude (90 days, up to 10 years with opt-in), Gemini (free version used for market research), and ChatGPT (free version retains data).
• As a safe alternative for students, James recommends Microsoft's Copilot for PowerPoint, which, at the time of the presentation, offered free access to GPT functionality within the Edge browser without data retention for model training.
• Mitigating LLM Hallucinations:
• One approach is to use LLMs to interrogate a user's own data (RAG), limiting the scope of potential errors.
• Local LLMs like FreedomGPT, GPT4All, and Ollama offer the benefit of running on personal devices, reducing privacy concerns and costs.
• Ollama allows users to create "model files" to tailor models for specific purposes, including adjusting parameters and setting system prompts to influence LLM behaviour.
• Image Generation with LLMs:
• James introduces the concepts of "prompts" (instructions), "recipes" (structured prompts), and "seeds" (starting points in the vector database) in the context of image generation.
• He recommends Adobe Creative Cloud for students interested in marketing careers as its model uses licensed data, reducing copyright concerns.
• While acknowledging limitations with branded terms and copyright, James encourages exploring prompt engineering techniques to refine image outputs.
• He highlights the ability to use seed values in Adobe Creative Cloud for creating consistent image variations.
• Emerging Trends: Video Generation and Prompt Engineering:
• James points to Adobe's upcoming video generation tool as a significant development, allowing users to generate videos from prompts.
• He emphasizes the importance of prompt engineering, offering a structured approach:
• Avoid "0 shot prompts" (simple questions) in favor of more detailed prompts.
• Employ "few shot prompts" (providing examples) to guide the model.
• Consider "chain of thought prompting" to encourage step-by-step reasoning.
• Utilize structured prompts that include context, instructions, input data, output indicators, and examples.
• Experiment with emotional manipulation and urgency in prompts.
• James advocates for the "ReAct" (Reason + Act) prompting structure, which involves the model clarifying the prompt and rating its own response, potentially reducing hallucinations.

Copyright, Privacy, and Ethics in AI:

• James stresses the importance of understanding copyright implications when using AI tools.
• He highlights the risk of LLMs inadvertently generating copyrighted content or images due to the datasets used for training.
• This raises concerns about potential infringement, particularly when using AI for creating logos or commercial images.
• He advises caution and awareness of ethical considerations when using AR devices like glasses for recording or identifying individuals without consent.

The AR Landscape and Future Trends:

• James provides a historical overview of AR, referencing the rise and fall of Google Glass, noting its initial promise and eventual market failure due to privacy concerns, high cost, lack of a "killer app," and negative public perception.

He analyses various current AR products, including:

• Smart glasses like Ray-Ban Stories, Echo Frames, and Vue Lite 2, noting their functionalities and limitations.

Niche AR applications like Raptor AR HUD for cyclists, Amazon's Room Decorator, and HoloLens 2 for industrial use.

Haptic feedback devices like the Wayfindr project and Humane's AI Pin, highlighting their experimental nature and limited market appeal.

He discusses the future of AR, predicting the dominance of Apple and Google in the hardware space and emphasizing the importance of low-cost, integrated devices.

Opportunities and Challenges in the AI Industry:

• James foresees a potential market crash in the AI industry due to oversaturation and unsustainable investment.
• He identifies DeepMind, led by Demis Hassabis, as a key player and influencer in the AI field.
• He advises students to focus on navigating the AI landscape, understanding available tools, and guiding companies in their AI adoption strategies.
• He acknowledges the high cost of developing LLMs (starting at $10 million) as a significant barrier to entry for startups and entrepreneurs.
• He predicts the future of AR will involve personalized, interactive, and location-aware experiences driven by AI, emphasizing the potential for location-based services and real-time data analysis.

Overall, the presentation provides a comprehensive and engaging overview of AI, LLMs, and AR, highlighting key concepts, trends, opportunities, and challenges. James emphasizes the importance of understanding the capabilities and limitations of these technologies, considering ethical implications, and adopting a critical and informed approach to their development and use.

What is the current state of artificial intelligence (AI)?
While AI has made significant strides, it is still primarily in the realm of narrow or weak AI, capable of completing specific tasks. Companies are striving towards artificial general intelligence (AGI), where AI can think and perceive like humans, but this hasn't been achieved yet. Artificial superintelligence (ASI), surpassing human intelligence, is a theoretical possibility for the future.

What are large language models (LLMs) and how do they work?
LLMs are AI systems trained on massive amounts of text data, allowing them to generate human-like text. They function by statistically predicting the next word in a sequence based on the input they receive. They don't understand the meaning or truth of the information they process, leading to potential inaccuracies and "hallucinations."

Are free AI tools safe for sensitive information?
Caution is advised when using free AI tools for sensitive information. Many free options retain user data for training purposes, potentially exposing your business plans or other confidential information. Consider paid versions or local models for enhanced privacy.

What are "foundation" and "frontier" models?
Foundation models are base models available for use and adaptation by other companies. Frontier models represent the most cutting-edge and advanced AI systems. Familiarising yourself with both types can enhance your job prospects.

What are some strategies for mitigating LLM hallucinations?
One method is using LLMs to analyse your own data, limiting the scope for errors. Another involves using structured prompts that clarify your request, ask for confirmation, and rate the results.

What are some ethical considerations regarding AI-generated content?
AI systems often train on copyrighted material, raising potential concerns about infringement. Ensure that your AI-generated content doesn't inadvertently violate copyright laws. Be mindful of privacy when using image generation tools, as they may create images of real individuals without their consent.

What is the future of augmented reality (AR)?
AR, which superimposes data onto the real world, is evolving beyond early iterations like Google Glass. Integration with everyday devices like glasses, watches, and phones is likely. However, challenges such as security, privacy, and compelling use cases need to be addressed for widespread adoption.

What are the career opportunities in the AI/AR landscape?
While hardware development is dominated by major players, there are opportunities in software development and AI/AR consulting. Understanding the available tools and their limitations, combined with creativity and ethical awareness, will be crucial for success.