Meta's future AI models to focus on consumer health
What's the story
Meta Platforms Inc.'s Chief AI Officer, Alexandr Wang, has revealed that the company's future artificial intelligence (AI) models will focus on consumer health. Speaking at the Bloomberg Tech conference in San Francisco, Wang stressed the importance of health in scaling these models. "Health is an area that we view as really critical as we scale these models out to billions," he said.
Career transition
Wang's background and role at Meta
Before joining Meta, Wang founded data-labeling start-up Scale AI. His transition to Meta came after CEO Mark Zuckerberg revamped the company's AI strategy, including a $14 billion investment in Scale. Now, Wang leads the newly-formed Meta Superintelligence Labs (MSL), which has already made some progress with its first AI model since Zuckerberg's multibillion-dollar AI overhaul.
Model launch
Muse Spark's performance and health capabilities
In April, MSL launched its first AI model under Wang's leadership, called Muse Spark. While it isn't as advanced as top models like Anthropic's Claude and OpenAI's ChatGPT, Wang said it's "better, frankly, than we expected internally." He also highlighted health capabilities as one of Muse Spark's biggest strengths and expects future models to be even more competitive.
Future plans
Integrating health features into consumer apps
Wang said as they build larger models, health will remain a focus area. He hinted at the possibility of integrating this feature into Meta's consumer-facing apps like Instagram, Facebook, and WhatsApp. This comes as AI companies race to build chatbots that appeal to consumers with health-related queries. However, this trend has also raised concerns over patient safety and privacy.
Risk assessment
Addressing biological risks and not open-sourcing Muse Spark
During Muse Spark's development, elevated biological-risk concerns were raised. Wang didn't elaborate on these risks but said Meta addressed them before the model's release. He also revealed that these findings influenced the decision not to make it open-source, a process that would have allowed outside developers access to the model's building blocks.