Meta debuts Muse Spark model with “complete overhaul” of AI

Meta released an AI model called Muse Spark on Wednesday. This represents a “first step” toward a “thorough overhaul of the system.” [its] AI initiatives. ”

Muse Spark is the first model from Meta Superintelligence Labs. Meta Superintelligence Labs was founded last year after CEO Mark Zuckerberg was frustrated with the progress of Meta and its Llama model and how it was lagging behind OpenAI’s ChatGPT and Anthropic’s Claude. Meta has hired former Scale AI co-founder and CEO Alexandr Wang to lead Meta Superintelligence Labs, investing $14.3 billion in the data label company for a 49% stake.

Now it’s time for Zuckerberg to see if his reconstituted AI team can captivate users.

Currently available on the web and in the Meta AI app, Muse Spark is expected to improve over time. The company plans to roll out a “contemplation” mode that can tackle more complex problems. Meta’s model uses multiple AI agents to work on the same problem at the same time, and says results are produced faster in deliberative mode.

“To increase inference testing time without significantly increasing latency, you can increase the number of parallel agents collaborating to solve difficult problems,” the company writes.

Meta’s competitors have traditionally kept these more sophisticated models behind paywalls. It’s unclear whether the meta will follow the same strategy.

But the company has jumped on one trend in the AI ​​industry. Meta said in a blog post that Muse Spark could be applied to help users solve health questions, something competitors are also working on.

Meta’s expansion into the health space, and even into the more basic functionality of logging into Muse Spark, may raise privacy concerns. Muse Spark users must log in with an existing Meta account, such as Facebook or Instagram, to use it. Meta does not explicitly state that personal information from Facebook and Instagram accounts will be used by the AI. But Meta is likely considering that it typically trains using public user data, and that the company is positioning Muse Spark as a personal superintelligence product.

Mehta also said Muse Spark performs especially well with visual STEM questions that can lead to “interactive experiences like creating fun mini-games or troubleshooting home appliances.”

In addition to investing in ScaleAI and hiring Wang, Zuckerberg’s company has also hired researchers from OpenAI, Anthropic, and Google. The bottom line is that if Meta is to become a true contender in the AI ​​industry, it’s now or never.

“Looking to the future, we plan to release increasingly advanced models that push the frontiers of intelligence and capabilities, including new open source models,” Zuckerberg wrote on Threads. “We’re building products that not only answer your questions, but act as agents that do things for you.”