Evaluating a travel-booking task
範例
Task to evaluate: "Plan a two-week trip and book the flights and hotel."
Givens from the sources: Muse can handle tasks like sending an email or booking travel; once a person shares a goal, Muse helps develop a personalized plan, coordinates time and resources, and advances the work on its own; it can open a browser and fill out forms; it checks with the person before sensitive actions like making a purchase; and Meta continues to invest in areas with current performance gaps, such as long-horizon agentic systems and coding workflows.
Assume nothing beyond these statements.
示範講解
Step 1 — Name the capability the task requires. The task needs planning, browser use, form filling, and a purchase. The sources name all four: planning a shared goal, opening a browser, filling out forms, and checking before a purchase. So the capability check passes.
Step 2 — Check the task shape against the named gaps. The named gaps are long-horizon agentic systems and coding workflows. A two-week trip is a bounded project with a clear end state, not an open-ended multi-day workflow, and it involves no code. So the gap check does not apply.
Step 3 — Check the approval boundary. The purchase is a sensitive action, so the source says Muse checks with the person first. That does not disqualify the use case; it changes what the person must be available for. The justified claim is that Muse can advance the booking and pause for approval before the purchase, not that it completes the purchase unattended.
Result
this is a justified use case, stated as "Muse can plan the trip, open a browser, fill forms, and advance the booking, pausing for approval before the purchase." The justification is the named capabilities plus the absence of a named gap for this task shape.
來源支援
- Introducing Muse: The World’s First Personal AI Agent Built for Everyone(about.fb.com · 已核准摘錄)
- How We Built Safety Into Muse | Meta AI Research(research.meta.ai · 已核准摘錄)