trending 📈

💰 Meta launches subscription. Is every social app becoming a subscription? Following in the footsteps of Snapchat+ and Twitter Blue, Meta Verified will allow users to finally get a blue checkmark on Facebook and Instagram for $11.99/month. Users will be required to submit a government ID, and will not be able to change their name, birthday, username or profile photo (!) without getting re-verified. CEO Mark Zuckerberg said the offering is about “increasing authenticity and security” - some analysts estimate it could also generate $2-$3B in annual revenue.

🤖 Meta releases LLM. In other Meta news, the company announced a new large language model called LLaMa (extra points for the name 🦙). How does this compare to OpenAI’s GPT-3? The model has far fewer parameters, making it faster and cheaper to deploy. It will be also made available to the AI research community, which is not (yet!) true of Google’s LaMDA and OpenAI’s ChatGPT - though other companies will not be able to use LLaMa for commercial purposes.

🪙 Coinbase reports earnings. It’s almost the end of earnings season. This week, Coinbase beat expectations on loss per share, at -$2.46 versus -$2.55 expected. However, the stock still fell as revenue declined sharply YoY, from $2.5B in Q4 2021 to $605M in Q4 2022. This is largely due to a decline in trading volume as the crypto market has cooled down - Coinbase’s non-trading revenue (from its staking business, interest income, and subscriptions) continued to climb.

Zain Kahn @heykahn

Spotify just announced DJ ― it's AI powered music recommender. It looks (and sounds) amazing 😍

Spotify is getting further into the AI game, with a new personalized DJ that delivers curated music recommendations and a commentator to introduce and talk about the tracks! It is currently rolling out in beta for a small set of Premium users.

It may have competition, as YouTube announced that users will now be able to build personalized radio stations that include recommendations of similar tracks/artists. We may finally be getting curation at scale 👀

what we’re following 👀

TikTok is coming to the Mercedes Benz in-car entertainment system.

How the best consumer marketplaces choose to display supply.

Artifact, a news app from the ex-Instagram founders, is now open to all - check it out!

This week, the U.S. Copyright Office revoked copyright protection for a comic book called “Zarya of the Dawn.” Why? They discovered that the images were created by an AI tool called Midjourney.

The author (Kris Kashtanova) and her lawyers argued that she followed a creative process similar to any artist using tools like a camera or Photoshop. While Midjourney technically generated the images, she wrote the prompts and iterated on the outputs hundreds of times to produce the final result.

An excerpt from the letter Kashtanova’s legal team sent to the USCO.

According to the USCO, this isn’t enough. They ruled that copyright is intended to protect humans, and that Midjourney was the “master mind” behind these images. The core of their argument is that users don’t have enough control over Midjourney’s output - it’s too unpredictable. Therefore, the images aren’t Kashtanova’s “own original mental conception.”

Kashtanova’s laywers believe that the USCO has a “factual misunderstanding” of how Midjourney works. The output is largely controlled by the prompt and iterations, and isn’t unpredictable. They argue that the images generated by Midjourney aren’t any more random than other types of art, like Jackson Pollock’s famous splatter paintings.

Text from https://processmechanics.com/2023/02/22/a-mixed-decision-from-the-us-copyright-office/, I added a Jackson Pollock image to compare!

The USCO decided that Kashtanova can keep the copyright on the text and layout of the comic, but not the images (though she may be filing a response that could change this outcome). We believe this is just the beginning of the debate around legal protections for AI-generated content. Some questions to ponder:

How much human input is required to qualify for copyright protection?
If Kashtanova used a tool that required more “work” (e.g. ControlNet), would the outcome be different?
How can the USCO prove if a piece of content was AI assisted or not?