Innovation Profs - 9/19/2023

Your weekly guide to generative AI tools and news

Generative AI News

Google nears release of AI software Gemini, The Information reports

Gemini, Google’s new conversational AI system, will soon be released as a competitor to GPT-4. A select number of companies have been given access to an early version of Gemini. Gemini features a collection of large language models and has text-to-image generation capabilities as well.

Uh oh, now AI is better than you at prompt engineering

Last week, a team of researchers at Google’s DeepMind posted a paper online describing OPRO, a program “that makes large language models try different prompts until they reach one that gets closest to solving a task.” The main idea behind OPRO is that it uses an algorithm called Meta-Prompt to find the optimal performance of various prompts as potential solutions to a given problem. The research team compared OPRO’s performance against human performance on several benchmarks, finding that the optimal prompts yielded better results than human-produced ones.

Did Google ban AI artists from running Stable Diffusion on its cloud?

Some members of the AI art community have claimed that they were banned from running Stable Diffusion on Google Colab, Google’s cloud computing platform. Chris Perry, the head of Colab, later clarified that the issue was the use of Gradio, a front-end interface of Stable Diffusion, with free accounts. As he tweeted out, “We operate the largest GPU subsidy on the planet, need to make hard choices sometimes.”

Americans are over generative AI like ChatGPT, but their bosses are just getting started

ChatGPT's global web traffic has declined for three straight months. The average time users spent on the platform declined as well. Is this the sign of a passing fad? (Note that the downturn took place over the summer months.). Employers sure don’t seem to think so, as a significant proportion of executives indicated in a survey that they plan to use generative AI to increase productivity.

Morgan Stanley kicks off generative AI era on Wall Street with assistant for financial advisors

Just yesterday, Morgan Stanley released its GPT-4 powered AI assistant, appropriately named “AI @ Morgan Stanley Assistant.” According to Morgan Stanley’s co-President Andy Saperstein, “Financial advisors will always be the center of Morgan Stanley wealth management’s universe. We also believe that generative AI will revolutionize client interactions, bring new efficiencies to advisor practices, and ultimately help free up time to do what you do best: serve your clients.”

These CMU researchers hacked ChatGPT. Now they’re targeting its brain for mind control.

A team of researchers at Carnegie Mellon have devised a new way to jailbreak a number of LLMs including ChatGPT, allowing them to bypass their content filters. Now, they are seeking to better understand the internal structure of LLMs, to determine how they make decisions and are sometimes prone to deceive users.

Generative AI Workshop Oct. 17

Our next Generative AI Workshop will be Oct. 17 at Drake University. Come learn how to use generative AI tools ChatGPT, Midjourney, D-ID and more. You can attend in-person or virtually. Sign up here.

Know someone who wants to learn more about generative AI? Forward them this newsletter or send them this link to sign up.

Tool of the week: Stable Audio

The company behind text-to-image tool Stable Diffusion released text-to-music AI generation tool Stable Audio.

Simply describe the audio you want with a text prompt, and the system generates it for you. Paid users can use the music in commercial projects and non-paying users can use the music in non-commercial projects.

Innovation Profs Homework

Adobe last week announced the launch of the new Firefly web application, a playground for exploring AI-assisted creative expression. Your generative AI homework this week is to give the tools a try.

The tools are free (for now at least), and include a text-to-image tool, a generative fill tool to remove objects or add new ones with text descriptions, a text-effects tool, and a generative recolor tool.

The text-to-image tool (seen below) lets you click to choose from popular styles.

If you make something, send it our way and we'll feature in next week's newsletter.

AI-generated image of the week

Just a reminder that AI-art generation tools still struggle with hands. The following set of images was created with Midjourney.

Prompt:  hand

Generative AI tip of the week

What do generative AI and storytelling have in common? A lot, actually. Storytelling is the backbone of good communication, and recently created a blog post on how to use generative AI for storytelling. The keys: start with an outline, nail your prompt and try different techniques. Read the full blog post.

If you haven’t yet tried, it’s a tool for creating presentations with generative AI.

Get starting with Generative AI

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

What we found

Take a look at these AI-generated images, which have generated a lot of buzz over the past few days.

Spiral village

Medieval village in checkerboard shape