Prueba de llamada a funciones paralelas

Las respuestas y explicaciones aparecen a medida que realiza el cuestionario.

1. En una llamada a función paralela, ¿qué devuelve el modelo en una respuesta?
   • The model trains on several datasets at once
   • The model requests several tool calls in a single response
   • Two models answer the same question
   • Tools call the model repeatedly
2. ¿Quién ejecuta realmente las herramientas al mismo tiempo?
   • The language model itself
   • The application or harness
   • The user's browser only
   • The model provider's billing system
3. ¿Qué par de llamadas NO se deben emitir en paralelo?
   • Weather lookups for three different cities
   • Searching two unrelated topics
   • Creating a folder and then writing a file into that folder
   • Fetching orders and support tickets for one customer
4. ¿Cómo se deben hacer coincidir los resultados con las llamadas a sus herramientas?
   • By the order they finish
   • By their position in the list
   • By the call identifier
   • By the length of each result
5. Dos de cinco búsquedas paralelas se agotan. ¿Cuál es la mejor respuesta?
   • Discard all five results
   • Return the three successes and two clear errors so the model can retry the failures
   • Crash the agent
   • Pretend the failed searches returned empty results without saying so
6. Why use Promise.allSettled or asyncio.gather with return_exceptions?
   • So one failed call does not discard the other results
   • To make calls run one at a time
   • To encrypt tool outputs
   • To reduce the number of tools
7. Roughly how long does a parallel batch of independent calls take?
   • The sum of all call durations
   • About the duration of the slowest call
   • Zero time
   • Always exactly one second
8. What helps prevent retries of side-effecting tools from duplicating actions?
   • Idempotency keys
   • Longer tool descriptions
   • Higher model temperature
   • Removing call identifiers