The new CEO of Character.AI — the controversial AI chatbot startup currently fighting a high-profile child welfare lawsuit over the suicide of a 14-year-old user — says the company has abandoned its founding mission of realizing artificial general intelligence, or AGI.
In an interview with Wired, recently crowned Character.AI CEO Karandeep Anand declared that the company "gave up" on the "aspirations" of its since-departed founders, Noam Shazeer and Daniel de Freitas.
It's a striking move for a young company that, by 2023, had reached a billion-dollar valuation while promoting a core mission to "bring personalized superintelligence to everyone on Earth."
"We are no longer doing that," Anand told Wired, emphasizing that the company has done "a lot of work" to shift away from building its own proprietary large language models (LLMs). The company now relies more than ever on open-source LLMs like Deepseek and Meta's Llama, said the CEO, with the executive adding that forsaking its AGI dreams has since allowed Character.AI to find "clarity and focus" around an "AI entertainment vision."
In many ways, the startup's whiplash-inducing pivots feel like an important pulse check on the AI industry and its many sweeping, fantastical promises, which many experts fear is a bubble waiting to pop.
The change in course, Wired notes, is backdropped by the cold reality that despite eye-popping investments from the likes of the venture capital firm Andreessen Horowitz and Google — which rehired Shazeer and de Freitas as part of a $2.7 billion acquihire last year — Character.AI has always struggled to actually generate revenue. That's a major problem, especially considering the process of building and training LLMs is extraordinarily expensive.
But while Character.AI might be saving some cash by relying on open-source LLMs, the shift away from making its own models, which started following the multibillion-dollar Google deal, continues to undercut the promise that made the company so allegedly valuable to its backers in the first place.
Character.AI and its investors once touted its enviable position as a "closed-loop" AI maker, highlighting the company's ability to continuously collect user inputs and feed them back into its model as training data.
"In a world where data is limited," Andreessen Horowitz partner and former Character.AI board member Sarah Wang wrote in a celebratory March 2023 blog post announcing the firm's high-dollar investment, "companies that can create a magical data feedback loop by connecting user engagement back into their underlying model to continuously improve their product will be among the biggest winners that emerge from this ecosystem."
Just over two years later, the situation looks drastically different. Character.AI hasn't just moved away from its original purported mission, but appears to be pretty far from the investor-praised value proposition that previously raked in billions.
... continue reading