Alexandr Wang, one of Meta’s most celebrated top officials and the company’s chief AI Officer, has hinted at a new strategy that suggests the company wants to pivot towards the health side of things.
In a recent conference held by Bloomberg, Wang revealed that Meta wants to make health advice and reasoning a cornerstone of the company’s long-term strategy, which could potentially give it an edge against the more enterprise-focused Anthropic and OpenAI.
“Health is an area that we view as really critical as we scale these models out to billions,” said Wang at the conference. With the company’s frontier AI models still in their infancy, the social media giant wants to stress that health is one of the key areas where the company can make a name for itself.
Muse Spark set Meta on course
Meta’s efforts at revamping its AI efforts bore fruit with the release of Muse Spark, a natively multimodal reasoning model. Although general-purpose benchmarks place the model slightly behind its peers in specialised tasks like coding, Wang expressed immense satisfaction with how the architecture handled early deployments.
He openly conceded that Muse Spark is “not at the tier of the leading frontier models” like Claude or ChatGPT, but he assured that the system performed “better, frankly, than we expected internally.”
Muse Spark is an early result of Meta’s MSL team, or Meta Superintelligence Labs, with the 29-year-old Wang at its lead. Wang, who was originally handling his own data-labelling startup, called Scale AI, was acquired by Mark Zuckerberg’s Meta for the role, and in doing so, became the company’s highest-paid and most influential executive. At the time, it was stated that Zuckerberg was pinning all his hopes and expectations on the MSL team to stay relevant and eventually lead the AI race.
Wang reveals why Meta chose to keep Muse Spark proprietary
So far, independent evaluations by the industry have shown that Meta’s Muse Spark is strongest in its medical and health-reasoning capabilities, outperforming several top frontier models on specialised medical benchmarks. Building a highly sophisticated health AI, however, has also brought unique safety and security hurdles. During the conference, Wang disclosed that Muse Spark had raised “biological-risk concerns” during its development phase.
While Wang did not elaborate on the specific nature of those hazards, he assured the audience that Meta successfully addressed and mitigated those risks before the model was rolled out to the public.
Wang also revealed that these biological-risk findings directly led to Meta not releasing Muse Spark as an open-source model, thus marking a pivot from the open-weights approach Meta popularised earlier with its Llama family. Wang said that keeping the underlying building blocks of Muse Spark proprietary allows Meta to maintain tighter safety governance over high-risk medical and biological data.
Wang, however, expected Meta’s future AI models to become more competitive as the development scales up.
