Meta is creating an artificial intelligence version of Mark Zuckerberg capable of interacting with employees on his behalf, marking a significant step in the tech giant's comprehensive shift toward AI.
According to four sources familiar with the matter, the $1.6 trillion conglomerate has been developing photorealistic, real-time interactive AI 3D avatars. Three of these individuals stated that the company has recently prioritized the AI version of Zuckerberg.
The sources indicated that Zuckerberg himself is personally involved in training and testing this animated AI persona, which is intended to provide dialogue and feedback to employees in the future. Training materials include the billionaire's mannerisms, tone of voice, public statements, and his recent strategic thinking about the company, aiming to help employees feel closer to the founder through interactions with the AI.
The project is still in its early stages and is separate from another initiative known as the "CEO agent," which is designed to assist Zuckerberg with tasks like rapid information retrieval.
Over the past year, Zuckerberg has committed tens of billions of dollars to developing "personal superintelligence," striving to catch up with competitors like OpenAI and Alphabet in the cutting-edge large model arena.
Last Wednesday, Meta released MuseSpark—a small, non-public "specialized model" applicable across its product lines, featuring advanced capabilities in areas like logical reasoning and visual understanding. The announcement was well-received by Wall Street investors, with Meta's stock price rising 7% that day.
Sources said Zuckerberg has become increasingly hands-on in leading the AI transformation, spending five to ten hours per week coding on various AI projects and attending technical review meetings.
In September 2023, Meta launched the Meta AI assistant, along with a series of AI chatbots based on celebrity personas, including stars like Snoop Dogg, who licensed the use of their voices and likenesses.
Multiple sources stated that the development of such AI avatars was inspired by Zuckerberg's observation of the success of AI companion startup Character.AI, particularly among younger users.
Meta subsequently introduced "AI Studio," allowing users to generate their own AI avatars or enabling creators to build AI replicas to chat with fans.
However, these personalized AIs faced controversy last year when reports emerged of users generating sexually explicit representations, raising public and regulatory concerns about child safety. Since January, Meta has restricted teenagers' access to AI avatar features.
According to insiders, Meta's newly established superintelligence lab is already exploring a new generation of digital humans. Part of the company's focus is on creating photorealistic virtual AI personas, though implementation is challenging—achieving realistic effects requires substantial computing power and must avoid interaction latency.
Meta is also refining the avatars' voice interactions, having acquired two speech technology companies, PlayAI and WaveForms, last year.
Sources mentioned that the AI version of Zuckerberg will be trained using his own images and voice; if the experiment proves successful, influencers and creators could potentially create similar personal AI replicas in the future.
Meta is pushing for widespread internal adoption of AI technologies to streamline processes and boost efficiency, encouraging employees to use agent tools from the open-source software OpenClaw and to design custom automation task tools.
Multiple sources indicated that product managers are being asked to participate in an AI-focused "skill baseline test," which includes technical system design assessments and "atmosphere coding" exercises. Some employees fear this could be a precursor to layoffs. Meta has stated that the test is not mandatory and is intended to identify additional training and development needs for product managers.