Innovation Profs - 2/18/2025

Your weekly guide to generative AI tools and news

In the Classroom with the Innovation Profs

Follow along as we teach Drake University’s first ever generative AI course.

This week’s topic: Intro to Text-to-Image

Lesson: This week’s class will be an introduction to generative AI image creation tools, how they work, how the technology has progressed, as well as the legal and ethical side of using them. Users generally have to pay to play in this space, but many tools offer a limited number of slower free image generations. We will start by exploring what tools can be used for free and comparing the quality of their outputs.

Some of our favorite image tools: Midjourney (no free option), Ideogram, Adobe Firefly

Homework: This week’s homework is about exploring what tools you can use for free and comparing the outputs they create. Your goal should be to find an image creation tool that you prefer to use. Here is the full of what our students will do.

Sign up for AI Lunch Clubs

We’re bringing back our generative AI lunch club. The next event will be an Introduction to Microsoft 365 Copilot on Feb. 26. If you are getting started with or considering adding Copilot for work, you won’t want to miss this one. We’ve been testing Copilot and find it an invaluable productivity tool.

More Lunch Clubs:
March 12: AI in Social Media 
March 26: AI Ethics
April 9: AI in Law and Policy

Latest Gen AI News

xAI releases Grok3

Elon Musk’s xAI has unveiled Grok-3, calling it “the smartest AI on Earth.” And early indications are that he is correct. An early version of Grok-3 ranks at the top of the Chatbot Arena leaderboard. The main Grok-3 model is being rolled out via the Grok app. A smaller Grok-3 mini version promises faster responses.

OpenAI postpones its o3 AI model in favor of a ‘unified’ next-gen release

If you’ve been waiting on the edge of your seat for the release of OpenAI’s reasoning model o3 after the launch of o3-mini and o3-mini-high earlier this year, you’re going to be somewhat disappointed. Last week OpenAI CEO Sam Altman announced that o3 will not be released as a standalone model but instead will be integrated into GPT-5, which will integrate many of the ChatGPT tools like canvas, search, and deep research. But prior to the launch of GPT-5, we should expect to see the release of GPT-4.5, code-named Orion, which according to Altman will be the last of OpenAI’s non-chain-of-thought reasoning models.

Scarlett Johansson calls for deepfake ban after AI video goes viral

Over the past week, you may have seen a video featuring the likenesses of a number of celebrities sporting a t-shirt that is both simultaneously pro-Jewish and anti-Kanye West. Unsurprisingly, the video is a deepfake. Now Scarlett Johansson, one of the celebrities whose likeness appears in the video, is pushing back with a call for a ban on deepfakes in the US: “It is terrifying that the U.S. government is paralyzed when it comes to passing legislation that protects all of its citizens against the imminent dangers of A.I.” This is not Johansson’s first high-profile run-in with AI, as she previously called out OpenAI for introducing ChatGPT’s voice mode with a voice that sounded an awful lot like her own voice (after she declined to allow OpenAI to use her voice in the product).

Anthropic prepares new Claude hybrid LLMs with reasoning capability

Avid readers of our newsletter likely recognize the horse race between OpenAI and Anthropic, with new developments with Claude being echoed in ChatGPT and vice versa (e.g. Claude artifacts vs ChatGPT with canvas, computer use with Claude vs OpenAI’s operator). It should come as no surprise that Anthropic has developed their own reasoning model to rival OpenAI’s gaggle of reasoning models. Slated for release in the coming month, Anthropic’s reasoning model will come equipped with variable resource allocation, which will allow users to vary how much computing power will be used to complete a given task.

Quick Hits

Tool of the week: Deepgram Nova-3

We’ve written about Deepgram in this space before. The company has great tools for understanding emotion in audio. Deepgram’s new Nova-3 is a cutting edge speech-to-text tool enhances transcription accuracy across diverse enterprise applications (such as ordering food in a noisy drive-thru). It introduces real-time multilingual transcription, supporting ten languages, and offers self-serve customization through Keyterm Prompting, allowing users to fine-tune up to 100 specific terms without retraining. Test Deepgram here.

AI-generated image of the week

It’s cold here in Iowa today. So cold that they delayed school so kid don’t have to be outside in the freezing temps. That made us think about our old friend the “make it more” generative AI photo trend. So we asked ChatGPT to make an image of kids bundled up in the cold, and then kept asking for more bundled up. Maybe the kids in the last photo are ready for school!

Prompt: kids bundled up on a cold iowa morning ... even more bundled up ... the most bundled up kids ever ...  even more

Get starting with Generative AI

New to generative AI? Here are some places to start…

What we found

Computer scientist Brian Hie developed Evo, a large language model trained on 2.7 million microbial genomes, enabling it to interpret and generate functional DNA sequences. This AI tool, akin to "ChatGPT for genomes," deciphers complex genetic patterns, potentially accelerating advancements in biological design and medicine.

And this…