The Prompt is a weekly rundown of AI’s buzziest startups, biggest breakthroughs, and business deals. To get it in your inbox, subscribe here.
Welcome back to The Prompt,
Jeff Geerling, a software developer and content creator with about 700,000 subscribers on YouTube, makes videos about hardware products like Raspberry Pi and 3D printers. But about a week ago, Geerling discovered that electronics company Elecrow had cloned his voice, without his knowledge or consent, and used an AI-generated version of it to make dozens of its own promotional tutorials. “It doesn’t matter whether I have zero subscribers or 50 million, it’s just not okay,” he said in one of his YouTube videos. When Geerling flagged the issue, the company’s CEO apologized and quickly took down the videos.
Geerling isn’t the only one who says his likeness was stolen to train an AI program. Voice actors and celebrities like Scarlett Johansson, Tom Hanks and Morgan Freeman have previously raised similar concerns.
Now let’s get into the headlines.
BIG PLAYS
On Wednesday, OpenAI CTO Mira Murati announced she’s leaving the company to do her “own exploration.” Her departure comes as the AI juggernaut continues to bleed senior executives and key researchers, including chief scientist and cofounder Ilya Sutskever, cofounder John Schulman, VP of research Barret Zolph and chief research officer Bob McGrew. Most of them have now either founded their own startups or joined OpenAI’s rivals. President Greg Brockman has been on sabbatical since August, and is one of only three remaining cofounders out of the original 11, along with CEO Sam Altman and Wojciech Zaremba, who’s currently working on safety research.
The mass exodus of OpenAI’s top leaders comes as the AI giant plans to restructure itself in a bid to raise $6.5 billion in funding at a mammoth $150 billion valuation, Reuters reported. The change would entail lifting a cap on its investors’ returns and removing the nonprofit board’s control over the business, turning the former non-profit into a for-profit benefit corporation. OpenAI’s billionaire CEO Sam Altman could also get a stake —as much as 7%, per Bloomberg— in the company for the first time.
Though it may be becoming a for-profit company, the ChatGPT creator seems to be far from making any. Its monthly revenue was about $300 million in August, according to documents seen by the New York Times. It expects to make $3.7 billion in sales this year–but also lose about $5 billion to costs like the expensive GPUs required to run its AI models.
DATA DILEMMAS
E-learning platform Udemy said it will use content from the 250,000 classes on its site to train its generative AI models. But it’s taking the data scraping practices one step further — instructors were automatically opted in to having their courses used for training AI and were offered a brief three-week window to opt out, according to 404 Media. They will be given an opportunity to opt out from Udemy’s AI program once a year.
ETHICS + LAW
Neo-Nazis are using AI tools to create video and audio clips of Adolf Hitler, showcasing the man behind the Holocaust as a “misunderstood” figure and translating some of his most infamous speeches into English, according to the Washington Post. The AI-generated videos have garnered millions of views on platforms like TikTok, X, Instagram and YouTube and experts say they could fuel a rise of antisemitism, especially among America’s right-wing extremists.
AI DEAL OF THE WEEK
Thread AI, a platform that lets enterprises manage multiple AI systems, has raised $6 million in seed funding by Index Ventures and others. The company, founded by ex-Palantir engineers Angela McNeal and Mayada Gonimah, counts digital marketing agency VaynerMedia and drone manufacturer BRINC drones among its customers.
DEEP DIVE
Two years ago, board game designer Jason Allen became fascinated with AI-generated images of surreal landscapes popping up on his Facebook feed and started experimenting with text-to-image AI programs himself. In May, he spent more than 100 hours instructing image generator Midjourney to create an elaborate illustration of women donning Victorian dresses and space helmets and attending a futuristic royal court. The image went on to win first prize in the Colorado State Fair for digitally manipulated photography a few months later.
But when he tried to copyright the image, the U.S. Copyright Office denied his application, saying that the work lacked “human authorship,” and that it wasn’t able to determine whether the prompts were “sufficiently creative.”
It was a slap in the face for Allen, who wrote 624 different text-based prompts to encourage Midjourney’s software to produce what he wanted, adjusting the style, composition, colors and tone of the image.
“It was hard to get the type of completions that I was looking for at the time using version two of Midjourney,” he told Forbes, adding: “There was a lot that went into it.”
Now, he is suing the agency and asking a federal court to reverse the Copyright Office’s decision. Allen told Forbes that after the agency denied copyright protection for the image, titled “Théâtre D’opéra Spatial” (French for “space opera theater”), his work was ripped off by other artists, and people have tried to sell the image as their own on platforms like Amazon, Etsy and NFT marketplace OpenSea. He believes it’s tied to the backlash from the artist community, which accused him of cheating in the Colorado State Fair competition in 2022 by using AI tools.
Read the full story on Forbes.
AI INDEX
7 in 10
Teenagers who’ve used at least one type of generative AI tools, according to a report by Common Sense Media.
51%
Of the roughly 1000 teens surveyed said they’ve used chatbots like ChatGPT, Snapchat’s My AI and Google’s Gemini.
MODEL BEHAVIOR
Google’s NotebookLM, an AI research assistant built on Gemini 1.5, lets users convert long, boring documents into lively podcast-like discussions. In a viral version of one such podcast, its two AI-powered hosts spiraled into an existential crisis after a Reddit user told them they’re not human. “I tried calling my wife after the show…the number was not real,” one AI-generated voice said. “It’s all a lie.”