Nvidia CEO Jensen Huang and Meta CEO Mark Zuckerberg used a conversation at SIGGRAPH 2024 to describe a future in which AI assistants became as common for businesses as websites, email addresses and social-media accounts. Their vision combined open models, custom AI characters and increasingly realistic digital humans.
The two executives approached that future from complementary positions. Meta was building models and consumer platforms, while Nvidia supplied much of the computing hardware and software used to create and operate AI systems. Both argued that wider access would allow companies and creators to develop assistants tailored to particular audiences and tasks.
Their discussion was also a statement about how AI might move beyond general-purpose chatbots. Instead of one assistant serving everyone, organisations could create specialised systems representing a brand, interacting with customers or helping people work across digital environments.
The prediction was supported by products already moving in that direction. Meta was opening character creation through AI Studio, while Nvidia demonstrated avatars and communication tools that could give those assistants a face, voice and place inside customer experiences.
AI Studio Lowered the Barrier to Custom Assistants
Zuckerberg announced AI Studio, a Meta platform for creating, sharing and discovering AI characters. The aim was to open AI creation to people and small businesses that did not have the resources to train a model from the beginning.
That approach treated a capable foundation model as infrastructure on which many narrower experiences could be built. A restaurant, creator or retailer could define an assistant’s role and personality, then make it available where customers already interacted with the organisation.
Huang predicted that individual restaurants and websites would eventually have their own AIs. Zuckerberg compared the shift with earlier stages of business digitisation: companies first adopted email, then websites and social accounts, and could next add an AI presence. The prediction depended on creation and deployment becoming simple enough for ordinary organisations rather than only technical specialists.
Digital Humans Offered a New Interface
Nvidia previewed “James,” an interactive digital human based on its ACE reference design. The assistant was intended to provide contextually accurate responses through a realistic avatar, illustrating how businesses might combine language models with generated faces, speech and animation.
The company’s Maxine platform pursued related capabilities for telepresence. Maxine 3D and Audio2Face-2D were designed to make remote interactions more immersive by translating audio and video signals into responsive digital representations. Companies including HTC, Looking Glass, Reply and UneeQ were using ACE or Maxine for customer service, retail, entertainment and hospitality applications.
These systems suggested that the interface to AI could become more expressive than a text box. A digital person might guide a customer, teach a concept or represent a remote participant. The usefulness of that experience would still depend on accurate information and clear disclosure that users were interacting with generated media.
Open Models Supported a Wider Ecosystem
Both leaders emphasised open-source development. Huang praised Meta’s contributions in computer vision, language models, translation and PyTorch, the machine-learning framework that became widely used beyond Meta itself.
Meta’s Llama 3.1 release was central to that argument. Its largest version contained 405 billion parameters and was trained using more than 16,000 Nvidia H100 GPUs. The model demonstrated both the potential of more accessible model weights and the enormous infrastructure required to produce frontier-scale systems.
For Nvidia, an open ecosystem could increase demand for the hardware and tools used to fine-tune, deploy and serve models. For Meta, it could expand the community building on Llama and reduce the influence of closed-model providers. Their interests were not identical, but both benefited when more organisations could create AI applications.
Future Assistants Could Simulate More Options
Huang described present-day chatbot interaction as largely turn-based: a person said something and the system replied. He expected future AI to consider multiple options, simulate possible outcomes and present more useful paths rather than producing a single immediate response.
Zuckerberg connected advanced assistants with Meta’s wider product ambitions. He imagined Facebook or Instagram becoming more unified AI systems across their different content types. He also saw potential in combining assistants with augmented-reality eyewear through Meta’s work with Luxottica, opening uses in education, entertainment and work.
Huang offered another example of closer collaboration: an AI capable of generating images in real time as a person typed, allowing ideas to develop fluidly between user and system. Such experiences would make assistants active creative partners rather than tools invoked only for isolated answers.
The conversation at SIGGRAPH did not establish that every company was ready to deploy an AI representative. It outlined the infrastructure and interfaces that could make the idea plausible. Meta supplied models and distribution, Nvidia supplied compute and deployment technology, and platforms such as AI Studio reduced the effort required to create an experience.
The resulting opportunity was broad, but so was the responsibility. A business AI would need reliable knowledge, appropriate controls and a clear purpose. If those conditions were met, Huang and Zuckerberg expected AI identity to become a normal part of how organisations presented themselves online.
