Our Reporter Tried Meta’s Viral AI Agent. Should You?
Summary
WSJ tech columnist Nicole Nguyen tests Meta’s new AI agent, Muse, which automates tasks like booking childcare and finding products. While the agent proved useful, it also made errors and aggressively requested access to sensitive personal data like emails and calendars. The episode highlights the trade-off between convenience and privacy, noting security risks like prompt injection attacks and Meta’s history. Nguyen ultimately decides to delete her agent, illustrating the polarizing nature of this emerging technology.
Topics discussed
Intro and announcement of live show in Dallas
Meta launches Muse AI agent and its capabilities
Comparison with ChatGPT and initial user hesitation
Show introduction and upcoming segments
Sponsorships: Intuit Credit Karma and Indeed
Naming the agent and customizing the avatar
Test 1: Finding and buying a bike rain cover
Test 2: Pilates referral error and AI hallucinations
Booking childcare and granting calendar access
Privacy concerns regarding sensitive personal data
Sponsorship: Gatorade Light
The trade-off: convenience vs. data access
Security risks: Prompt injection and hacking
Meta's security measures and trust issues
Public opinion and impact on online advertising
Future of agentic AI and data portability
Final verdict: Deleting the agent data
Outro and post-show banter
Listen ad-free on Castria