Meta Unveils Personal Assistant AI 'Muse'

The U.S. Treasury Department announced its Treasury bond buyback plan. Kim Dante mentioned, "The Secretary of the Treasury announced a buyback of Treasury bonds," which is interpreted as a measure to counter the decline in long-term Treasury bond prices.

However, the announced scale was $6 billion, falling short of market expectations of $7-8 billion. Immediately after the announcement, the 10-year Treasury yield rose, reaching a 52-week high, reflecting market disappointment.

Major semiconductor-related companies such as Micron, AMD, Intel, and SanDisk saw their stock prices strengthen. Kim Dante stated, "Today, Micron, AMD, Intel, SanDisk, these guys are doing well. Why are they doing well? The ripple effect of GPT-6 Astra is still in the market."

This is due to the continued market expectation that the emergence of GPT-6 Astra will lead to an expansion of artificial intelligence (AI)-related capital expenditures (CapEx).

Kim Dante stated, "Muse's initial usage today far exceeded our expectations."

Initial usage significantly surpassed Meta's predictions, with a utilization rate 10 times higher than that of the user test group, drawing significant market attention.

Kim Dante explained Muse's core function, saying, "It's my assistant that remembers everything I've said about myself."

Muse provides proxy execution capabilities, such as directly accessing websites, making reservations, and processing payments on behalf of the user, based on individual virtual machines.

The AI agent Muse operates by setting up a separate virtual machine, a dedicated computer, for each user. Kim Dante explained, "To implement functions like booking and reservations, a separate computer is set up for each user."

This virtual machine directly logs in using the user's ID and password, performs clicks on their behalf, and is linked with the Stripe payment system, enabling actual payments.

Muse offers active monitoring and information analysis functions for users. Kim Dante explained the feature by giving a hypothetical example: "Hey, monitoring shows that tickets suddenly dropped by $40 on Friday, making them cheaper."

Specifically, it constantly monitors flight ticket price fluctuations, offering notifications for the lowest prices and suggesting purchases. It also provides personalized services through a user-customized instruction command system.

Meta shared a real user case for Muse. Kim Dante mentioned one case, stating, "Muse found over $1,000 in uncollected checks for me." This case demonstrates that Muse provided a tangible benefit to the user by recovering over $1,000 in uncollected checks.

AI agents perform various practical tasks such as flight bookings, used car transactions, and negotiating communication fees. Kim Dante stated, "It can book flights, sell used cars, negotiate phone bills, and it keeps running even if you turn off your phone."

Even if the user turns off their phone or is offline, it operates independently on a cloud basis, continuing its tasks.

Kim Dante explained, "The great point is that it keeps studying at night. So, the user keeps providing information, right?"

Through this learning, Muse provides proactive, customized services such as creating game guides or suggesting history study content.

Kim Dante stated, "Since ordinary people can't do it one by one, Meta's radical decision is to say, 'We will run everything directly on the cloud for you.'"

Previously, users faced significant initial burdens, such as having to purchase a Mac computer. However, Meta has increased accessibility by implementing it as a consumer product running directly on its own cloud.

This is analyzed as an intention to ensure cost efficiency by utilizing its own lightweight model, 'Muse Spark 1.3,' and to allow more users to experience AI agents.

Kim Dante explained the revenue model, stating, "Using Muse will earn you money and save you money. The agent will take a small commission on transactions it facilitates."

Initially, 100 million tokens per week will be provided for free, but Meta plans to introduce premium user subscriptions, such as 'Power' and 'Maximum,' to offer additional features.

Meta's strategy is to generate revenue not through traditional ad impressions, but by taking a small commission from actual transactions mediated by the AI agent.

Reuters reported that Meta had been tracking how its employees used their computers. Kim Dante conveyed this news, saying, "There was news that Meta had actually been tracking everything about how its employees use computers."

This data is presumed to have been used to train AI in controlling computers and performing actual tasks.

Meta is attempting to change the paradigm of the existing advertising market. Kim Dante explained Meta's changed strategy, stating, "The traditional advertising market was a market that sold impressions and clicks. Meta wants to change that entire paradigm. So it proposes things to me first."

This aims to evolve from merely inducing ad impressions and clicks to an AI agent proactively understanding user patterns and suggesting products directly for search and purchase when needed.

Through personalized AI, Meta intends to anticipate user needs, and by having the agent directly mediate purchases, maximize advertising effectiveness and create new markets.