Discover more from Artificial Ignorance
AI Roundup 011: Another viral AI moment (feat. Drake)
April 21, 2023
Won't someone think of the celebrities?
If you weren't paying attention, you might have missed the wild ride that was "Heart On My Sleeve," a song by @ghostwriter977.
Released Friday, the song used cloned voices of Drake and The Weeknd, the most authentic sounding to date.
By Tuesday, it had millions of streams on Spotify and YouTube.
By Thursday, it was gone - due to Universal Music Group's copyright claims.
Why it matters:
UMG claims the song is a "violation of copyright law," but it's a gray area. Voices generally aren't copyright-able, and there's certainly no precedent for AI-generated voices.
Celebrities are joining the millions of artists, writers, and programmers affected by AI. If AI can perfectly mimic your content - what does that mean for you?
No such thing as a free lunch
Large language models like ChatGPT train on billions of documents, many from the internet. Now, the internet wants its cut. Both Reddit and StackOverflow plan to charge AI companies to use their data.
The big picture:
This isn't a new trend. But AI is becoming a potential cash-grab for social platforms. Or at least another reason to restrict user-generated content.
Restrictions (and regulations) tend to benefit incumbents. OpenAI can afford to pay Reddit and StackOverflow. Scrappy startups, not so much.
But will it matter? We're trying to shrink the amount of training data needed for LLMs, with mixed results so far. It's possible we reach a checkpoint where social media data is unnecessary.
Sam Altman, for what it's worth, says that the "age of giant AI models is already over."
Thanks for reading! Subscribe for free to get updates right in your inbox.
Google gets going
After weeks of Microsoft stealing the headlines, Google made plenty of news this week - but not all of it was positive.
CEO Sundar Pichai has been on an interview spree. This week, he told 60 Minutes that AI "is going to impact every product across every company."
Disgruntled Google employees have also been on an interview spree. Leaked internal chats show workers calling Bard "a pathological liar" and "worse than useless."
Stability AI announces its first language model, StableLM. Microsoft is testing a new in-house AI chip. Elon Musk founds X.AI to work on “TruthGPT.” (so much for that AI pause). Atlassian adds its own AI assistant. World Photography Award winner rejects prize, reveals AI-generated submission.