AI Today and Everyday — April 17, 2024

What’s new in the world of AI

Meta in hot water over explicit AI images

Our Report: Meta’s independent Oversight Board is investigating Meta’s policies surrounding explicit, AI-generated content, following two incidents where AI-generated, explicit images were posted to Facebook and Instagram, which Meta’s systems failed to detect and respond to.

Key Points:

• In both incidents, Meta has removed the content, but with the first, it failed to respond to a user (who reported the explicit content twice), before it was escalated to the Board.
• In the second, Meta detected an explicit AI image of a US public figure and took it down, raising questions over why it removed the first image, but not the second.
• The board will investigate both cases — they’re asking for public comments (until April 30th) — and will post its decision over Meta’s handling of each case, in a few weeks.

Why you should care: Meta has recently announced plans to widen the scope of content it reviews and labels with “Made with AI”, but these two incidents suggest an inconsistent enforcement policy regarding AI-generated deep fakes, which explains why the Board is stepping in to investigate.

Ex-OpenAI Director: “OpenAI needs to be audited”

• Ex-OpenAI Director (Helen Toner) has called for AI leaders — such as Google, Microsoft, and OpenAI — to be audited, so “others can examine the information they provide.”
• She’s insisting they share information about “what their systems can do and how they’re managing risks”, claiming there needs to be a reporting system, for when AI goes wrong.
• Toner supported the ousting of OpenAI CEO, Sam Altman (after she co-authored a paper criticizing its safety practices), and resigned when the decision was reversed.

Adobe Considers Third-Party Tools for Premiere Pro

Adobe Premiere Pro is set for an AI-powered revamp.

The software is exploring integration with third-party AI tools, including collaborations with OpenAI, to offer features like object insertion and distraction removal, making video editing even more efficient.

Here’s a recently released demo from Adobe featuring the AI Generative Fill feature on video:

Generative AI in Premiere Pro powered by Adobe Firefly | Adobe Video

While Adobe maintains its proprietary Firefly AI model, this move acknowledges the evolving landscape with competitors like OpenAI, Midjourney, and Pika Labs. Read more

Apple iOS 18 Will Be On-Device (1 minute read)

Apple’s upcoming AI features in iOS 18 are rumored to focus on privacy, with the initial set of enhancements functioning entirely on-device without the need for an internet connection or cloud-based processing, thanks to the company’s in-house large language model known internally as “Ajax.”

Google’s New Technique Gives LLMs Infinite Context (5 minute read)

Google researchers have introduced Infini-attention, a technique that enables LLMs to work with text of infinite length while keeping memory and compute requirements constant.

Google ads get new AI features

• Google announced generative AI features for its advertising platform, Demand Gen, which accelerates the creation of Google ads across platforms like YouTube and Gmail.
• The new features allow advertisers to create images — tailored to their branding needs — using text prompts, and use a “generate more like this” tool to re-create high-performing assets.
• To comply with safety policies, AI images will include markings — like invisible digital watermarks — and the features are available now, in English, with more languages coming this year.

Amazon copies Spotify with AI playlists

• Amazon Music is testing a new AI-powered playlist generator, “Maestro” (similar to Spotify’s AI playlist feature) where users can generate playlists using spoken, text, and even emoji prompts.
• Users can write or speak prompts (or choose from a list of prompt suggestions) that include activities, sounds, or emotions, and an AI-generated playlist will appear.
• While Maestro is in beta for Amazon Music users and Prime and Amazon Music subscribers in the US, Amazon has
• warned that it “won’t always get it right the first time.”