Welcome back to The Prompt,
Hustle culture that involves long hours and late nights has always been a fixture in Silicon Valley. But in the race to ship products and features faster, some AI startups are encouraging–and in some cases, requiring–employees to work 6 or 7 days a week, Forbes exclusively reported. Among them are some of the buzziest like hiring platform Mercor and customer service-focused tool Decagon.
Now let’s get into the headlines.
BIG PLAYS
OpenAI’s complicated restructuring endeavors have taken yet another turn. The company announced that its nonprofit parent will continue to have control over ChatGPT and other products, a reversal of an earlier announcement that it would not. OpenAI plans to go ahead with converting its structure into a public benefit corporation to continue raising capital and remove caps on investors’ returns. The decision comes after it faced staunch opposition from former cofounder and rival Elon Musk, as well as a number of former employees, who claimed that OpenAI was straying away from its original mission.
Back in 2015, OpenAI was founded as a nonprofit research lab, but the recent popularity of ChatGPT sparked an AI arms race in Silicon Valley that has resulted in more demand for its tools. To meet that demand, OpenAI announced a $40 billion investment from Softbank and others, $30 billion of which is contingent on the company’s restructuring.
AI DEALS OF THE WEEK
Identity verification startup Persona raised $200 million in Series D funding led by Ribbit Capital and Founders Fund, reaching a $2 billion valuation. The company is helping companies OpenAI, LinkedIn, Reddit and DoorDash ensure that their users are real and who they say they are— an increasingly challenging task as billions of bot surf the internet with the rise of AI agents.
Decagon, which is building AI agents for customer service, is in talks to raise $100 million at a $1.5 billion valuation, Forbes exclusively reported.
Cybersecurity startup Doppel is developing AI agents to automatically flag hundreds of malicious ads and fake accounts for businesses like Notion, and has raised $35 million in new funding led by Bessemer Venture Partners.
OpenAI has agreed to buy AI-powered coding tool Windsurf (formerly Codeium) for $3 billion, Bloomberg reported. The company was featured on the 2024 Forbes Next Billion Dollar Startup List and the Forbes AI 50 list this year. The news comes as Cursor, one of the hottest AI tools of the year, has closed a $900 million round at a $9 billion valuation, the Financial Times reported.
DEEP DIVE
Entrepreneurs often create products to cater to niche problems in a specific industry. But for Stack AI cofounder and CEO Bernard Aceituno, the opposite is also true.
His company is creating chatbots and AI agents that can automate a broad buffet of back office jobs across functions like finance, HR and customer support, doing things like processing RFPs, reviewing contracts, conducting audits and helping train staff. Today, its users have created some 100,000 agents with it, Aceituno claims. These are already in use within some 200 organizations including Nubank as well as one defense agency (Aceituno did not specify which.) Local government agencies in California, Texas and Massachusetts are using Stack’s artificial intelligence systems to pour through piles of documents and extract data that can then be used for a bunch of other tasks, “cutting middle management bottlenecks,” Aceituno said.
Aceituno claims his company’s AI agents are helping healthcare companies with constrained budgets and staff crunches, deal with thousands of inquiries. Government contracts are using them to fill out dozens of applications with just 20% of the workforces.
The company announced Tuesday it has raised $16 million in Series A funding led by Lobby Capital.
Aceituno, who previously built visual-based machine learning models at Meta, says “AI models are trapped behind Jupiter notebooks.” With Stack AI, enterprises can connect their agents to their internal documents and databases and use off the shelf model from over 30 providers.
Originally from Venezuela, Aceituno first experimented with GPT 3 to summarize parts of his own PhD thesis at MIT. He believes tech stacks should be more versatile and applicable to a wide host of tasks. He claims his company offers a platform with the building blocks that can be customized based on the task at hand, while taking data security and privacy into account.
“We want to be the Excels of the world,” he says.
MODEL BEHAVIOR
AI companies have deployed a slew of so-called “reasoning” models that appear to break down problems into a series of logical steps to produce an answer. But it turns out these advanced systems are more prone to generating errors and making up false information (also known as hallucination), than the previous generation of models, according to The New York Times. For newer AI models, hallucination rates can reach as high as 79%.
