AI

AI’s Multimodal Future Is Here. Integrating New AI Capabilities In The Classroom.

AI image generated using Google ImageFX from a prompt “Create an image of a professor training an AI computer chip as if it was a dog in a university classroom.” https://labs.google/fx/tools/image-fx
~ K. Quesenberry ~ Leave a comment

In my last post, I needed a pep talk. In teaching digital and social media marketing I’m used to scrambling to keep up with innovations. But AI is a whole other pace. It’s as if I’m trying to keep up with Usain Bolt when I’m used to running marathons.

Like the marathon I signed up for in July, November comes quickly. No matter how training goes the start time comes, the horn goes off, and you run. Here comes the Spring semester. No matter the number of AI updates dropped in December I need to show up ready to go in early January.

If I want to make a difference and have an influence on how AI impacts my discipline and teaching, I don’t have a choice. I can relate to what AI expert Ethan Molick said in his latest Substack,

“This isn’t steady progress – we’re watching AI take uneven leaps past our ability to easily gauge its implications. And this suggests that the opportunity to shape how these technologies transform your field exists now when the situation is fluid, and not after the transformation is complete.”

The other morning, when I should’ve been finishing Fall grades, I spent a couple of hours exploring AI updates and planning how I’ll advance AI integration for Spring. Instead of AI bans (illustrated by the Fahrenheit 451 inspired image of my last post), I’m going deeper with how we can train AI to be our teaching friend, not foe.

AI image generated using Google ImageFX from a prompt “Create an image of a professor training an AI computer chip as if it was a dog in a university classroom.” https://labs.google/fx/tools/image-fx
AI image generated using Google ImageFX from a prompt “Create an image of a professor training an AI computer chip as if it was a dog in a university classroom.” https://labs.google/fx/tools/image-fx

NotebookLM opens up teaching possibilities.

A lot of new AI updates came this Fall. One that caught my eye was Google’s NotebookLM. In a NotebookLM post, I explained how I was blown away by its Audio Overview of my academic research that it turned into an engaging podcast of two hosts explaining the implications for social media managers.

I see potential to integrate it into my Spring Digital Marketing course. NotebookLM is described as a virtual research assistant –  an AI tool to help you explore and take notes about a source or sources that you upload. Each project you work on is saved in a Notebook that you title.

These are the various notebooks I’ve used so far for research and the new course notebook.
The various notebooks I’ve used so far for research and for my Digital Marketing class.

Whatever reference you upload or link, NotebookLM becomes an expert on that information. It uses your sources to answer questions and complete requests. Responses include clickable citations that take you to where the information came from in sources.

As a Google Workspace for Education user, uploads, queries, and responses are not reviewed by human reviewers or used to train AI models. If you use your personal Google account and choose to provide feedback, human reviewers may see what you submit. To learn more click here.

Source files can be Google Docs, Google Slides, PDFs, Text files, Web URLs, Copy-pasted text, public YouTube video URLs, and Audio files. Each can contain up to 500,000 words or 200MB files. Each notebook can contain up to 50 sources. Added up NotebookLM’s context window is large compared to other models. ChatGPT 4o’s context window is roughly 96,000 words.

When you upload to NotebookLM, it creates an overview summarizing sources, key topics, and suggested questions. It also has a set of standard documents with an FAQ, Study Guide, Table of Contents, Timeline, or Briefing Doc. An impressive feature is the Audio Overview which generates an audio file of two podcast hosts explaining your source or sources.

NotebookLM as an AI tutor.

I plan on using NotebookLM as an AI tutor for students in my Spring Digital Marketing course. I like the open-source text I’ve been using for years, but the author has stopped updates. The strategic process and concepts are sound, so I update content with outside reading and in-class instruction.

I tested NotebookLM creating a notebook for Digital Marketing course resources. First, I uploaded the PDF of the text. Then, I added website links to six digital marketing websites that I use for assigned readings and in-class teaching. Finally, I added my blog. I plan to show students how to create theirs at the beginning of the semester.

This is my notebook for Digital Marketing. I was impressed with asking it questions that I often get from students about assignments.
This is my notebook for Digital Marketing. I was impressed with the answers it gave to questions I often get from students.

AI may not be accurate 100% of the time, but controlling the sources seems to help and puts less pressure on crafting a perfect prompt. My discipline knowledge knows when it gets something wrong. I tested my Digital Marketing NotebookLM asking questions on how to complete main course assignments such as personal branding blogs, email, SEO, and content audits. I haven’t noticed any wrong answers thus far.

Important note about copyright.

I’m testing NotebookLM in this class because my main text is open source and all the websites I link to are publicly published sites (not behind paywalls). Google is clear about its copyright policy,

“Do not share copyrighted content without authorization or provide links to sites where people can obtain unauthorized downloads of copyrighted content.”

We should set a good example and educate students by not uploading copyrighted books or information only accessible through subscriptions or library databases. Below is my general AI policy for the course.

The policy carves out acceptable and helpful uses of AI while explaining the ways AI should not be used.
This policy carves out acceptable/helpful AI use while explaining ways AI shouldn’t be used.

In completing final reports students will access information behind paywalls such as Mintel reports. They’ll add the information and cite it as they’ve done in the past. The goal isn’t to use NotebookLM to complete their assignments for them. The goal is to give them a resource to better understand how to complete their assignments.

NotebookLM as a study tool.

I see NotebookLM as a positive tool for student learning if used as a study guide, reinforcement, or tutor. It would have a negative impact if used to simply replace reading and listening. What’s missed when you use AI in the wrong way is depicted in an infographic I created for a previous blog post on the importance of subject matter expertise when using AI.

For a website assignment, my course NotebookLM gave a nice summary of the process and best practices to follow. That’s something students often struggle to find in the text and other sources. The assignment requires pulling from multiple chapters and resources. The notebook summary included direct links to the information from various text chapters and digital marketing blogs. I also tested its accuracy with questions about an email assignment and had it create a useful study guide.

This will be so helpful for an assignment that student often miss steps and best practices as it draws from multiple parts of the text.
Answering questions will be helpful in assignments where students often miss steps and best practices that draw from multiple parts of the text and readings.

Students can create audio overviews of podcast hosts talking about a topic drawing from the sources. Impressively, when I asked for an Audio Overview explaining the value of a personal professional blog assignment to students it understood the student’s perspective of thinking blogs are outdated. It began, “As a student, I know you’re thinking blogs are outdated, but personal professional blogs are a great …” The Audio Overview also adjusted the text process for businesses and applied it to a personal branding perspective.

Going beyond Copilot in other areas.

I also plan on students leveraging new AI capabilities in Adobe Express and Google’s ImageFX in multiple classes. Our students have free access to Adobe Creative Suite where new AI capabilities go beyond Firefly generated images. In Express you can give it text prompts to create mockups of Instagram and Facebook posts, Instagram stories, YouTube thumbnails, etc.

Students' ideas will be able to be expressed even better with Abobe’s new text to create AI interface in Adobe Express along with its image creation capabilities with Firefly.
Students’ ideas can be expressed better with the text to create AI interface in Adobe Express along with the image creation capabilities of Firefly.

AI’s multimodal future is here.

That other morning I also dove deeper into new AI multimodal capabilities. It was so remarkable I recorded videos of my experience. I explored new live audio interactions in NotebookLM and created a demonstration of what’s possible with Google’s Gemini 2.0 multimodal live video.

I was blown away when testing the new ability to “Join” the conversation of the podcast hosts in NotebookLM’s Audio Overview. While the hosts explained the value of a personal professional blog, I interrupted asking questions with my voice.

 

Near the beginning, the hosts tell students to write about their unique skills. I clicked a “Join” button and they said something like,

“Looks like someone wants to talk.” I asked, “How do you know your unique skills?” They said “Good question,” gave good tips, and continued with the main subject. Later I interrupted and asked, “Can you summarize what you have covered so far?” They said sure, gave a nice summary, and then picked back up where they left off.

Finally, I interrupted to ask a common student question, “What if I’m nervous about publishing a public blog?” The hosts reassured me saying people value honesty and personality, not perfection. What really impressed me was the hosts answering questions about things not specifically in the sources. They could apply concepts from the sources to understand the unique perspective of a given audience.

Multimodal AI as a live co-worker.

This last demonstration of the new multimodal capabilities of AI is for my own use. With Gemini 2.0 in my Google AI Studio account, I could interact in real time using text, voice, video, or screen sharing.

The video below is a demonstration of what’s possible in live video and conversations with Gemini 2.0 as it “sees” what‘s on my screen. I had a conversation with it to get feedback on the outline for my new five-part AI integration workshop I’m planning this Spring for faculty on campus.

Writing the last two blog posts was time well spent.

Planning what I’ll do in the Spring and writing these last two blog posts has taken me two-three days. Because it was 100% human created there was a struggle and a time commitment. But that is how I learn. This knowledge is in my memory so I can explain it, apply it, or answer questions.

Talking to Gemini was helpful, but it doesn’t compare to the conversations I’ve had with colleagues. AI doesn’t know what it feels like to be a professor, professional, or human in this unprecedented moment. Let me know how you’re moving beyond AI bans and where you’re executing caution.

I have a lot of work to do to implement these ideas. That starting horn for the new semester is approaching fast.

100% Human Created!

AI Task Framework: Examples of What I’d Outsource To AI And What I Wouldn’t.

Copilot created this image of a college age man sitting said in a basement looking lonely at an old dusty unused exercise bike.
~ K. Quesenberry ~ Leave a comment

This is the second post in a series of five on AI. In my last post, I introduced an AI task framework to be more intentional about why and how we use AI in our jobs, businesses, or organizations. In this post, I give examples based on my previous advertising career.

AI Framework Template for AI Use Click on the image to download a PDF template.

As an advertising copywriter, some everyday Tasks and Goals included:

  1. Fill out timesheets detailing what I worked on each day to bill time to clients and projects to get paid.
  2. Research a client’s business and industry to demonstrate knowledge of their unique challenges and opportunities.
  3. Create ideas for campaigns and individual ads to sell to a client and publish to meet marketing objectives.
  4. Write social media ad copy for social media marketing to generate engagement and conversions for a client.

(1.) I would outsource timesheets to AI.

I envision an AI assistant that Extracts (AI Function) file use logs from programs like Microsoft Word, Categorizes (AI Function) by job number, and Creates (Level of Thinking) a spreadsheet listing client, job, and time. I could review and adjust it before submitting.

After thinking of this example, I discovered that Microsoft is adding this capability. Copilot for time entry creates time entries for team members without navigating through forms or filling details with dropdowns, generating first drafts for users to modify and confirm for timesheet submission.

The Level of Thinking in this example is Applying a process to Create a suggestion for my time entry (AI Capabilities). It doesn’t require creativity or imagination and I maintain final human judgment on accuracy (Distinctive Human Skill). By tracking job numbers no Copyrighted or Proprietary data is used. Human impact is positive. Everyone I knew hated timesheets. We loved coming up with ideas (Legal & Ethical Use).

(2.) AI could help with some aspects of client research.

AI could Answer Questions (AI Function) like “What are the current challenges and opportunities in the ice cream industry?” An open system like GPT would give me general answers based on open sources from the internet that may or may not be the most current, accurate, or relevant.

AI is Understanding (Level of Thinking) on a cursory level (AI Capability). To contextualize this understanding to your client and judge for accuracy (Distinctive Human Skill) you need proprietary data from paid databases like Mintel, your client, or your own research. Your personal experience with the client or industry is an added Distinctive Human Skill.

You could outsource this to AI by uploading proprietary data into an AI model Summarize and Ask Questions. (AI Function). But you’re uploading Copyrighted/Proprietary material without permission (Legal & Ethical Use). Mintel forbids input into AI systems and clients are adding AI restrictions to contracts to protect their data from training LLM models a competitor could use.

Some are developing Closed AI versus Open AI systems that run locally storing data on their computers versus the cloud. The ad/PR agency network Publicis is investing in an internal AI built on proprietary data. When available this could be a great way to quickly get up to speed on a business and industry.

How much I’d outsource depends on my previous experience. If it was a new client or market I was unfamiliar with I may worry how much I’d Understand (Level of Thinking) or Remember (Level of Thinking) if AI did it all. In an in-person meeting could I recall or contextualize the information on the fly?

(3.) AI could help with some parts of idea generation.

I would outsource some brainstorming to AI, not idea formation, but AI could give me more material for ideas by Answering Questions (AI Function). Let’s say my client wants to sell water bottles to 25-34-year-olds. I could ask “What do 25-34-year-olds who work out look for in a water bottle?” and “What are current trends with 25-34-year-olds who work out?”

With these prompts, GPT via Copilot Created (Level of Thinking) a list of alternatives (AI capability). From the list, I put together a feature “one-hand operation” with a trend of “functional fitness.” Then I Asked for functional fitness examples. From that list, I put together a humorous image or video scene of a young woman easily sipping out of her Owala water bottle with one hand while swinging a heavy Kettlebell with the other. This formulated an original solution (Distinctive Human Skill).

Evaluating AI responses and knowing what to Ask (Level of Thinking) comes from knowledge of the client, problem, market, target, and trends to discern the best and identify AI hallucinations. I’d also use my domain expertise of what concepts are good Remembering (Level of Thinking) from my long-term memory of 17 years of creating ideas for clients (Distinctive Human Skill).

I wouldn’t have AI write ad copy or scripts directly. If it isn’t mostly Created by a human, it can’t be copyrighted to sell to your client or to protect them from use by competitors (Legal & Ethical Use). I’d also check my agency and client for specific restrictions on AI. Your Knowledge (Level of Thinking) of the client and humans (Distinctive Human Skill) is better at Creating (Level of Thinking) less generic more human copy and scripts.

(4.) AI could help in parts of social media campaign creation.

AI could help brainstorm content Answering (AI Function) “What kind of content do 25-34-year-olds who work out like to see on social media?” I’d Evaluate (Level of Thinking) AI’s best suggestions (Distinctive Human Skill). One was “personal anecdotes.” It reminded me of an insight I read in a Mintel report about unused home workout equipment.

I combine this with the text “Peloton brings the motivation of a community to your home.” This gave me a visual idea of unused home workout equipment. I could mockup the social idea using AI to Generate (AI Function) the image. I’d ask “Create an image of an unused, dusty, stationary bike in a basement with a lonely looking guy” (Level of Thinking). This image would help me sell the idea to the client.

Generated with AI (DALL-E 3 via Copilot Designer ∙ June 25, 2024 at 1:33 PM

After approval, my art director and I would consider Copyright issues. Using AI-created artwork for commercial use is unsettled due to sources for training data. Adobe Firefly claimed to be copyright-compliant, but revelations about training data may put Firefly users at legal risk. A trusted photographer may be best to ensure compliance (Legal & Ethical Use).

We’d also consider that the medium sends a message. Does an artificial human and image support Petoton’s message of genuine human connection? I’d weigh the risk of uncanny valley. When tech gets too close to human people get an unsettled feeling. That creepy feeling can be transferred into negative feelings about the brand. Toys R Us and Under Armour have faced backlash for using AI generated video in this way. Google sparked backlash over an ad where a dad had AI write a letter for his daughter because it had to be perfect (Legal & Ethical Use).

I can’t help thinking about the human impact. I’ve worked with many talented creators who add to my ideas with their expertise. If we all decide to use AI instead, photographers, models, illustrators, designers, and writers lose their livelihoods. Levi’s faced a backlash after announcing they’d use AI generated models (Legal & Ethical Use).

Creating content variations (AI Capabilities) is a tedious part of social media. AI could help Generate (AI Function) variations to fit different platforms. I could ask “Write this copy ‘Peloton brings the motivation and community of a gym to the convenience of your home’ in 10 different ways.” I could also tell it to write a specific length for each platform’s character limits. This type of AI outsourcing is happening. Meta Ad Manager is adding Text Variations and social media management software Hootsuite has OwlyWriter AI.

Going through this AI task exercise makes me hopeful.

Breaking down my job into tasks making intentional decisions on what to outsource to AI gives me hope. It reminds me of our human agency. It helps me visualize what Mollick describes in his book Co-Intelligence. Instead of replacing all human tasks, we can use AI as Centaur (division of tasks) and Cyborg (intertwined alternating subtasks).

Once you decide what tasks to outsource you need to know how to ask AI to get the best results. In my next post, I’ll dive deeper into prompt writing.

This Was Human Created Content!