AI Agent Functionality and User Interaction

What is the main purpose of the unified agent?

• To perform simple, repetitive tasks quickly.

• To specialize in a single type of task, like deep research.

• To seamlessly transition between thinking, taking actions, and using various tools for complex tasks.

• To only produce visual outputs like slides and images.

Which of the following are types of tools accessible to the unified agent?

• Text browser, visual browser, and terminal.

• Spreadsheet editor, presentation software, and image generator.

• Email client, video conferencing software, and chat applications.

• Gaming console, music player, and social media platform.

How does the agent learn to select the most appropriate tool for a given task?

• Through pre-programmed decision trees.

• By receiving explicit instructions from the user for each step.

• By being rewarded for correctly and efficiently solving problems during training.

• By randomly trying different tools until one works.

What is a key complementary strength between Deep Research and Operator?

• Deep Research excels at processing long articles, while Operator is better at interacting with web pages.

• Deep Research is better at interacting with web pages, while Operator excels at processing long articles.

• Both Deep Research and Operator excel at the same tasks, making them redundant.

• Deep Research and Operator are not complementary and serve entirely different purposes.

What was a highly requested feature for Deep Research based on customer feedback?

• The ability to log in to websites and access authenticated sources.

• Improved visual interaction with web pages.

• Faster processing of short articles.

• Better integration with email services.

What is a key capability of the Agent model regarding user interaction during long tasks?

• It can only receive instructions at the beginning of a task.

• It allows users to interrupt and provide new guidance or clarification.

• It automatically corrects all user errors without input.

• It requires users to restart the task if new information is needed.

What measures are in place to mitigate risks when using AI agents, and what user action is recommended for sensitive information?

• Models are trained to ignore suspicious instructions, and users should always share sensitive information directly with the agent.

• Layers of monitors oversee agent actions, and users should use "takeover mode" for direct input of sensitive data.

• Agents are programmed to never ask for sensitive information, and real-time updates prevent all new attacks.

• Users are encouraged to share all information freely, as the system is foolproof against all attacks.

Which of the following tasks can an AI agent perform based on the demonstrations?

• Ordering custom stickers, creating a wedding preparation report, and planning an MLB stadium itinerary.

• Designing complex 3D models, composing original music, and performing advanced medical diagnoses.

• Managing personal finances, driving autonomous vehicles, and conducting scientific experiments.

• Translating languages in real-time, generating realistic video footage, and building physical robots.