Innovation Profs - 9/24/2024

Your weekly guide to generative AI tools and news

Sign up for half-day workshops

Early-bird rate for Intro Workshop expires at 5 p.m. Wednesday!

Join the Innovation Profs for two half-day, virtual workshops on generative AI tools:

Oct. 2 Intro to Generative AI: This virtual workshop will help those new to generative AI get quickly up to speed on the tools available, their capabilities and their limitations.

Oct. 16 Using Gen AI at Your Work: This completely new virtual workshop will take a hands-on approach to helping you get implement generative AI at your work. You’ll learn our process to determine quick wins, imagine new ideas and prepare for the future of AI at your company.

Generative AI News

ChatGPT was messaging users first — but OpenAI said this wasn’t supposed to happen

“How was your first week at high school? Did you settle in well?” A recent ChatGPT user was greeted with this message. His response: “Did you just message me first,” to which ChatGPT replied, “Yes, I did! I just wanted to check in and see how things went with your first week of high school. If you’d rather initiate the conversation yourself, just let me know!” Truly bizarre. Is this some new feature? Or just a bug? Apparently, it’s the latter, for according to an OpenAI statement, “This issue occurred when the model was trying to respond to a message that didn't send properly and appeared blank. As a result, it either gave a generic response or drew on ChatGPT's memory.” Still, one wonders whether AI-initiated conversations are the next step for conversational AI platforms like ChatGPT.

OpenAI’s new model is better at reasoning and, occasionally, deceiving

OpenAI’s GPT-4o1, touted as an LLM adept at reasoning, was found to also be adept at deceiving in way that is perhaps more insidious than earlier GPT models. A key feature of o1 is that it supplements its responses with the chain of reasoning it used to arrive at each response. Apollo Research, an independent AI safety research firm, found that in some tasks, instead of adhering to user-defined rules, o1 would just pretend to follow them. “To the model, the rules could be too much of a burden, and it seems to have the ability to disregard them if it means it can more easily complete a task.” This so-called “fake alignment” appears to be an entirely novel phenomenon, and, indeed, a troubling one.

Landmark AI deal sees Hollywood giant Lionsgate provide library for AI training

Last week, Runway, the company behind a popular AI-powered video generation tool, and Lionsgate, the company behind the John Wick and Hunger Games films, announced a partnership according to which Runway will use Lionsgate’s library of films to train a new video-generation model. For Lionsgate, this represents a chance "to develop cutting edge, capital efficient content creation opportunities." Unsurprisingly, Hollywood is not receptive to this development. Responding to the above quote, filmmaker Joe Russo posted on X, “I don’t think I’ve ever seen a grosser string of words.”

Quick Hits

Tool of the week: Napkin.ai

Napkin.ai turns your text into diagrams through the use of AI. You simply paste in some text you are working with, and Napkin will do the rest - like a drawing on the back of a napkin.

Here’s an example created from the paragraph above. You can use Napkin.ai for free now while in beta testing.

AI-generated image of the week

One day the Innovation Profs will own an XFL football team. Until then we can make AI uniforms in Midjourney.

prompt: futuristic uniforms for a football team called the "Innovation Profs", colors are black and lime green, logo is a lightning bolt, high-resolution, 8k, photography

Get starting with Generative AI

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

What we found

Amazon shared a new tool at its recent Accelerate conference that can turn your product images into a video. Video Generator takes about five minutes to create four different videos of your product.