Welcome back to The Prompt,
Sam Altman’s cryptocurrency venture, Worldcoin, which scans the irises of people from around the world in exchange for a payment in crypto, is rebranding itself to just “World.” The company has shifted its strategic focus away from crypto and into human identification in a world where AI is omnipresent. Now people can scan their eyes to verify that they are in fact humans and not bots, my colleague Rich Nieva reported.
Now let’s get into the headlines.
BIG PLAYS
OpenAI CEO Sam Altman once touted the AI juggernaut’s relationship with Microsoft as the “best bromance in tech.” A new report from the New York Times suggests the relationship between the two companies is deteriorating. For example, employees at the two companies are at odds over how much computing power OpenAI needs to be given to build the most cutting edge technologies. Additionally, Microsoft has forged ties with OpenAI’s competitors like Inflection, whose CEO Mustafa Suleyman now leads Microsoft’s consumer AI efforts.
ETHICS + LAW
Dow Jones, the parent company of the Wall Street Journal, and the New York Post are suing AI search startup Perplexity, alleging “massive” copyright infringement, Variety reported. According to the lawsuit, the news companies sent a letter to Perplexity in July 2024 outlining the legal implications of scrapping its articles without consent and offered a licensing deal but the company never responded. The lawsuit alleges Perplexity helps people “skip the links” to the original sources of information, diverting customers and revenues away to its own search platform by copying copyrighted information.
The lawsuit comes after Forbes found in June that the buzzy startup, now reportedly raising funding on a $8 billion valuation, had republished entire sections of paywalled articles and pushed them out to its users as a podcast, blog and a YouTube video.
AI DEAL OF THE WEEK
Billionaire-backed nonprofit Earth Species Project has raised $17 million to build an AI system to help decode animal communication, Forbes reported. The organization is backed by entities like Linkedin cofounder Reid Hoffman and the Paul G. Allen Family Foundation and is building a model that analyzes patterns in vocalizations of animals like beluga, crows and zebra finches. Its end goal is to be able to communicate with animals–or at least understand what they are saying.
DEEP DIVE
If you look at AI-generated videos today, most of them look like live photos rather than real videos, said Paras Jain, CEO of AI video startup Genmo. That’s in part because video generation AI models don’t understand the laws of physics and aren’t trained on examples of different real world actions and complicated movements.
One example of a movement that is particularly challenging for AI to duplicate is gymnastics: “We know what gymnastics is supposed to look like conceptually. But then the model needs to understand a great deal of physics to get that right. It needs to know that a person swinging is like a double pendulum,” Genmo cofounder Ajay Jain said.
Brothers and PhD graduates from UC Berkeley, Ajay Jain and Paras Jain hope that their new open source video generation model called Mochi 1, can change that. With about $29 million in funding from backers like NEA, the duo are building AI models that can generate high-definition videos with better motion quality— ones that make different scenes in a video look more fluid.
Genmo is operating in a vastly cluttered space with companies big (like Meta, which recently launched Meta Gen, a family of models for video generation) and small (like Pika Labs, whose latest model can crush, melt and turn anything into cake). “I think that there’s an upcoming Cambrian explosion of AI video applications. But my hope is that we’ll empower developers to take our models, build products on top of them, and then share the innovations with the community,” Jain told Forbes in an interview.
The startup, valued at $160 million and with about 2 million users, also hopes that its new models will be able to better adhere to prompts that are fed into it, Jain said. “I was looking at early artist experiences with Sora and one aspect an artist called out was it was hard to prompt,” he said.
WEEKLY DEMO
Writing generated by AI tools is often wordy, robotic and isn’t well structured, Forbes contributor Laura Brown writes. “While using GenAI might be saving your time, it’s likely also wasting your reader’s,” she writes. Instead, she suggests writing conversationally and simply and also recommends considering whether using an AI tool—which may have to be prompted multiple times to get the best output— is even needed in the first place.
AI INDEX
Hugging Face, an online library for hundreds of thousands of open source AI models, has hosted thousands of malicious models with code that can poison data and steal information, security researchers told Forbes. Fake profiles of accounts that claim to be companies like Meta, Visa and SapeX are also swarming the platform.
3,000
Malicious models that researchers have found on Hugging Face this year.
QUIZ
This tech giant is being sued by a film production company for allegedly feeding images from sci-fi film “Blade Runner 2049” into an AI generator to create similar images for promoting its product.
- Tesla
- Apple
- Microsoft
Check if you got it right.
MODEL BEHAVIOR
One of the world’s largest book publishers, Penguin Random House, has amended the copyright wording imprinted on each book to forbid anyone from using the book or any parts of it to train artificial intelligence systems, The Bookseller reported. The change comes after a band of authors, artists and content creators have sued AI companies for using their copyrighted works without permission to train their models.
