Innovation Profs - 6/18/2024

Your weekly guide to generative AI tools and news

Generative AI News

Microsoft’s all-knowing Recall AI feature is being delayed

Back in late May we reported on Microsoft’s new AI-powered Copilot+ PCs, which were planned to include Recall, a tool allowing users to search any of the actions they have previously performed on their computer. Apparently, the way Recall obtains this information is by taking screenshots of everything users do on their computers. Unsurprisingly, a number of cybersecurity concerns about the feature have been raised. In light of these concerns, Recall is now being delayed and will not be available when Microsoft’s new Copilot+ PCs are released today. A modified version of the tool will be tested as part of Microsoft’s Windows Insider program before being more widely released.

OpenAI expands lobbying team to influence regulation

OpenAI’s global affairs team is growing from three members to thirty-five, with the aim of reaching fifty members by the end of 2024. This team of lobbyists will work as consultants on AI legislation across the globe. According to Anna Makanju, VP of government affairs at OpenAI, “We are not approaching this from a perspective of we just need to get in there and quash regulations . . . because we don’t have a goal of maximising profit; we have a goal of making sure that AGI benefits all of humanity.”

Adobe’s AI Tools in Photoshop and Lightroom To Be Limited by ‘Generative Credits’

Adobe users will have their use of Firefly-powered tools constrained by what Adobe is referring to as “generative credits.” Apparently, running out of credits doesn’t mean that users will not be able to access Firefly without obtaining new credits. According to Adobe, “After the plan-specific number of generative credits is reached, you can keep taking generative AI actions to create vector graphics or standard-resolution images, but your use of those generative AI features may be slower.”

How Anthropic’s comprehensive red team methods close AI security gaps

Last week, Anthropic released their guidelines to red-teaming its AI models (i.e., attacking its own models to determine security vulnerabilities), which are more comprehensive than those published by competitors OpenAI, Google, and Microsoft, as well as NVIDIA and NIST. In a blog post, Anthropic outlines four methods of red-teaming: domain-specific expert red teaming, using language models to red team, red teaming in new modalities, and open-ended general red teaming. Anthropic is taking a human-in-the-middle approach to red-teaming, focusing on human insights and intuition to guide the process of making their models more secure.

Quick Hits

Tool of the week: Stable Diffusion 3

Stability AI has released Stable Diffusion 3 Medium, which it calls its “most advanced text-to-image open model yet.”

SD3 Medium promises photorealistic results without complex workflows.

It also is reported to solve the three most common issues with generative AI images - hands, faces and text.

Stability called the model's ability to generate text “unprecedented.”

There's not free model of this, but plans start at $9 per month.

Innovation Profs Homework

Make an AI video. There are several players in the generative video space, and we want you to try out one of the currently available tools. Those include Luma’s Dream Machine, Runway’s Gen-2 and PikaLabs. Give one a try to see the current state of generative AI video.

AI-generated image of the week

We asked Meta AI to create a photo that represents summer in Iowa. Then I asked it to do the same thing for our neighbor to the west, Nebraska. Can you tell which one is supposed to be Iowa?

prompt: Summer in Iowa / Nebraska

Generative AI tip of the week

TikTok is going all-in on AI tools for creators. This week, the social network announced new tools as part of its Symphony advertising tools. TikTok announced Symphony Digital Avatars, AI Dubbing for global translations, and our Symphony Collective to help creators and brands produce TikTok-first content that breaks through.

Symphony is currently in beta testing. You can join the waitlist here.

Get starting with Generative AI

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

What we found

Runway has announced its latest text-to-video model Gen-3 Alpha.

This comes on the heals of four other major generative video models - OpenAI’s Sora, Luma’s Dream Machine, Google’s Veo and Kling AI from Chinese tech giant Kuaishou Technology.

Runway calls Gen-3 a major improvement in fidelity, consistency, and motion over Gen-2. The new model isn’t available to use yet, but you can try out Gen-2.