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.
Scarlett Johansson’s lawyers have demanded that OpenAI take down a voice for ChatGPT that sounds much like her own after she’d declined to work with the company to create it. The actress said in a statement provided to Forbes that her lawyers have asked the AI company to detail the “exact processes” it used to create the voice, which sounds eerily similar to Johansson’s voiceover work in the sci-fi movie Her. “I was shocked, angered and in disbelief,” she said.
The actress said in the statement that last September Sam Altman offered to hire her to voice ChatGPT, adding that her voice would be comforting to people. She turned down the offer, citing personal reasons. Two days before OpenAI launched its latest model, GPT-4o, Altman reached out again, asking her to reconsider. But before she could respond, the voice was used in a demo, where it flirted, laughed and sang on stage. (“Oh stop it! You’re making me blush,” the voice said to the employee presenting the demo.)
On Monday, OpenAI said it would take down the voice, while claiming that it is not “an imitation of Scarlett Johansson” and that it had partnered with professional voice actors to create it. But Altman’s one-word tweet – “Her” – posted after the demo last week only further fueled the connection between the AI’s voice and Johannson’s.
Now, let’s get into the headlines.
BIG PLAYS
Google made a long string of AI-related announcements at its annual developer conference last week. The biggest one is that AI overviews — AI-generated summaries on any topic that will sit on top of search results — are rolling out to everyone across the U.S. But users were quick to express their frustration with the inaccuracies of these AI-generated snapshots. “90% of the results are pure nonsense or just incorrect,” one person wrote. “I literally might just stop using Google if I can't figure out how to turn off the damn AI overview,” another posted on X.
Consumers will also be able to use videos recorded with Google Lens to search for answers to questions like “What breed is this dog?” or “How do I fix this?” Plus, a new feature built on Gemini models will let them search their Google Photos gallery. Workspace products are getting an AI uplift as well: Google’s AI model Gemini 1.5 will let paying users find and summarize information in their Google Drive, Docs, Slides, Sheets and Gmail, and help generate content across these apps. Meanwhile, Google hired artists like actor and filmmaker Donald Glover and musician Wyclef Jean to promote Google’s new video and music creation AI tools.
DeepMind CEO Demis Hassabis touted Project Astra, a “universal assistant” that the company claims can see, hear and speak while understanding its surroundings. In a demo, the multimodel AI agent helps identify and fix pieces of code, create a band name and even find misplaced glasses.
TALENT RESHUFFLE
Key safety researchers at OpenAI, including cofounder and Chief Scientist Ilya Sutskever and machine learning researcher Jan Leike, have resigned. The two led the company’s efforts to develop ways to control AI systems that might become smarter than humans and prevent them from going rogue at the company’s superalignment team, which now no longer exists, according to Wired. In a thread on X, Leike wrote: “Over the past few months my team has been sailing against the wind. Sometimes we were struggling for compute and it was getting harder and harder to get this crucial research done. Over the past years, safety culture and processes have taken a backseat to shiny products.”
The departure of these researchers also shone a light on OpenAI’s strict and binding nondisclosure agreements and off-boarding documents. Employees who refused to sign them when they left the company risked losing their vested equity in the company, according to Vox. OpenAI CEO Sam Altman responded on X saying “there was a provision about potential equity cancellation in our previous exit docs; although we never clawed anything back, it should never have been something we had in any documents or communication.”
AI DEALS OF THE WEEK
Scale AI has raised $1 billion at a $14 billion valuation in a round led by Accel. Amazon, Meta, Intel Capital and AMD Ventures are among the firm’s new investors. The company has hired hundreds of thousands of contractors in countries like Kenya and Venezuela through its in-house agency RemoTasks to complete data labeling tasks for training AI models, Forbes reported last year. In February, Forbes reported that the startup secretly scrapped a deal with TikTok amid national security concerns.
Plus: Coactive AI, which sorts through and structures a company’s visual data, has raised a $30 million round at a $200 million valuation led by Emerson Collective and Cherryrock Capital. And London-based PolyAI, which sells generative AI voice assistants for customer service and was cofounded by three machine learning PhD students at Cambridge, has raised $50 million at a nearly $500 million valuation.
DEEP DIVE
The girls in the photos on TikTok and Instagram look like they could be 5 or 6 years old. On the older end, not quite 13. They’re pictured in lace and leather, bikinis and crop tops. They’re dressed suggestively as nurses, superheroes, ballerinas and french maids. Some wear bunny ears or devil horns; others, pigtails and oversized glasses. They’re black, white and Asian, blondes, redheads and brunettes. They were all made with AI, and they’ve become magnets for the attention of a troubling audience on some of the biggest social media apps in the world—older men.
“AI makes great works of art: I would like to have a pretty little virgin like that in my hands to make it mine,” one TikTok user commented on a recent post of young blonde girls in maid outfits, with bows around their necks and flowers in their hair.
Similar remarks flooded photos of AI kids on Instagram. “I would love to take her innocence even if she’s a fake image,” one person wrote on a post of a small, pale child dressed as a bride. On another, of a young girl in short-shorts, the same user commented on “her cute pair of small size [breasts],” depicted as two apple emojis, “and her perfect innocent slice of cherry pie down below.”
Forbes found hundreds of posts and comments like these on images of AI-generated kids on the platforms from 2024 alone. Many were tagged to musical hits—like Beyonce’s “Texas Hold ‘Em,” Taylor Swift’s “Shake It Off” and Tracy Chapman’s “Fast Car”—to help them reach more eyeballs.
Child predators have prowled most every major social media app—where they can hide behind screens and anonymous usernames—but TikTok and Instagram’s popularity with teens and minors has made them both top destinations. And though platforms’ struggle to crack down on child sexual abuse material (or CSAM) predates today’s AI boom, AI text-to-image generators are making it even easier for predators to find or create exactly what they’re looking for.
TikTok and Instagram permanently removed the accounts, videos and comments referenced in this story after Forbes asked about them; both companies said they violated platform rules.
Read the full story in Forbes here.
YOUR WEEKLY DEMO
On Monday, Microsoft introduced a new line of Windows computers that have a suite of AI features built-in. Called “Copilot+ PCs”, the computers come equipped with AI-powered apps deployed locally on the device so you can run them without using an internet connection. The computers can record your screen to help you find anything you may have seen on it, generate images from text-based prompts and translate audio from 40 languages. Sold by brands like Dell, Lenovo and Samsung, the computers are able to do all this without internet access because their Qualcomm Snapdragon chips have a dedicated AI processor. The company claims its new laptops are about 60% faster and have 20% more battery life than Apple’s MacBook Air M3, and the first models will be on sale in mid-June.
MODEL BEHAVIOR
In the past, universities have invited esteemed alumni to deliver commencement speeches at graduation ceremonies. This year, some institutions turned to AI. At D’Youville University in Buffalo, New York, a rather creepy-looking robot named Sophia delivered the commencement speech, doling out generic life lessons to an audience of 2,000 people. At Rensselaer Polytechnic Institute’s bicentennial graduation ceremony, GPT-4 was used to generate a speech from the perspective of Emily Warren Roebling, who helped complete the construction of the Brooklyn Bridge and received a posthumous degree from the university. The speech was read out by actress Liz Wisan.
