Meta Platforms introduced its latest artificial intelligence model on Wednesday. This marks the first model released since CEO Mark Zuckerberg reorganized the company's AI division with a multi-billion dollar investment to keep pace with competitors.
The highly anticipated model, named Muse Spark, was developed by the Meta Superintelligence Labs (MSL). This new team, led by Chief AI Officer Alexandr Wang, brings together highly-compensated AI researchers. Muse Spark will power the Meta AI chatbot, representing a shift from the company's previous open-source strategy to a closed-source model, meaning its design and code will not be publicly available.
Following the announcement, Meta's stock price surged 6% in New York.
This model is the first major test for Zuckerberg's newly established MSL lab. After a series of setbacks, he expressed dissatisfaction with the company's inability to keep up with rivals like OpenAI, Anthropic PBC, and Alphabet Inc.'s Google. Last year, he made a $14 billion investment in Scale AI, bringing Wang on board. Since then, Zuckerberg has invested billions in AI talent and committed hundreds of billions more to infrastructure, such as data centers, to catch up.
A company executive stated that Meta has been trying to keep its AI unit agile, granting researchers greater autonomy and streamlining the previously management-heavy organizational structure. The executive, who spoke anonymously due to discussing internal matters, mentioned that Wang has approximately 100 direct reports.
The executive acknowledged that Muse Spark's performance in some areas lags behind OpenAI's ChatGPT, Anthropic's Claude, or Google's Gemini, but stated that the company is still in the early stages of execution. In a blog post, Meta described the model as "an early data point on our trajectory," noting that several larger models are currently in development.
Muse Spark, which had the internal codename "Avocado" during its nine-month development, is viewed by executives as an update to Meta's AI strategy, which previously focused on the open-source Llama models. Wang supports closed-source models, and although Meta still plans to develop open-source models in the future, the executive said the company is also considering selling API access to Muse Spark.
The executive indicated that the Meta AI chatbot will remain free for users, but the company is considering a future subscription-based pricing model.
Muse Spark utilized several third-party open-source models during its training process, including Alibaba's Tongyi Qianwen, as well as models from OpenAI and Google.
"Like others in the industry, Meta uses techniques such as distillation to learn from publicly available AI models, with strict safety guardrails in place, to improve our own models," a Meta spokesperson said.
Comments