Daily Digest: Detecting AI images

PLUS: YouTube's AI bet

February 07, 2024

Hello folks, here’s what we have today;

PICKS

  1. Both OpenAI and Meta are making AI images easy to spot:

    • Images generated in ChatGPT and OpenAI’s API will include metadata using C2PA specifications.🍿Our Summary (also below)

    • Meta’s labeling AI-generated images on Facebook, Instagram and Threads.🍿Our Summary (also below)

    Google has been using its Synth ID tech to add this info via its AI image tools.

  2. YouTube CEO’s 4 big bets for 2024 have AI on the top. Neal Mohan, YouTube’s CEO is focusing heavily on AI, creator empowerment, and expanding YouTube's presence in our living rooms. YouTube is betting big on AI to spark a new wave of creativity while doubling down on protecting creators.🍿Our Summary

  3. The abundance agenda by a16z - What does a16z, one of the top VC firms, think about AI and the upcoming transformation in consumer technology? Well, it’s telling us upfront.🍿Our Summary

TOP TOOLS

  • Galileo 1.0 - The prompt-to-UI platform, now open to all. (disclaimer: I’m an investor)

  • Imprompt AI - Chat with everything with the control of an Agent Mesh.

  • V-Day song by Suno - Make a song for your Valentine. (It’s good, I tested, maybe less focus on the name though)

  • Coze - Quickly create bots without coding and publish them.

  • Kompas AI - Sophisticated AI agents at work.

  • Geppetto - Open source AI companion for your Slack teams.

  • Metavoice 1B - TTS with emotions and zero-shot cloning.

  • Artbox - A library of high-quality visuals for your next project.

NEWS
QUICK BITES

OpenAI's stepping up its game by embedding C2PA magic into images made by DALL·E 3. That means any pic you get from ChatGPT or the API comes with an ID card you can check out.

What is going on here?

Images generated via ChatGPT and OpenAI API now include metadata using C2PA specifications.

What does this mean?

C2PA is a techie standard that's all about making sure the media we see and share is legit. Think of it as a digital ID card for images, telling you where they came from and what's their story.

So, every image you get from DALL·E 3 now comes with this C2PA tag - it's like a seal of authenticity. But here's the kicker: not everything's foolproof. This tag can get lost in the digital shuffle, like when images hit social media or get screenshot. OpenAI's clear that while C2PA's cool, it's not the end-all for proving where an image came from.

Why should I care?

Knowing what's real and what's not is getting tougher. There’s a flood of AI images coming that we’ll have to deal with every day (most of us already are). Adding C2PA tags on AI images is a try to sift through that noise, even if it's not perfect.

QUICK BITES

Meta's stamping "Imagined with AI" labels on AI-generated images across Facebook, Instagram, and Threads. The goal? Making sure we know when we're looking at something a machine cooked up versus a human.

What is going on here?

Meta's rolling out labels for AI-generated images, aiming for transparency.

What does this mean?

The folks at Meta are not new to the AI game, but now they're pushing for transparency by labelling AI-crafted pics. They've been tagging their own AI creations with "Imagined with AI" tags, but they're looking to expand this transparency across the board. They're in talks with industry pals to create a common language for AI content. Meta says whether you're scrolling through dog pics or deep dives into distant galaxies, if it's AI-generated, you should know.

Meta’s also stepping up its game with invisible watermarks and metadata to make this process slicker, not just on its platforms but across the web. But much of this is removable with rather simple options (like taking a screenshot in many cases).

For audio and video, they’re making a feature for self-labelling AI content and if Meta finds out, you haven’t followed through, it’s gonna release the algorithm dogs on you (the bad ones).

Why should I care?

Imagine scrolling and not knowing if what you see is someone's beach photo or AI's dream vacation. These are early days for the spread of AI-generated content, so any indication, even if removable helps filter AI content.

There also has to be human caution in identifying AI-generated stuff, which is very hard. So look out for signals away from the content: is this info likely to be true? Does the person sharing this often talk gibberish and trust fake news?

Ben’s Bites Insights

We have 2 databases that are updated daily which you can access by sharing Ben’s Bites using the link below;

  • All 10k+ links we’ve covered, easily filterable (1 referral)

  • 6k+ AI company funding rounds from Jan 2022, including investors, amounts, stage etc (3 referrals)

Reply

or to participate.