GCP-GAIL 試験練習問題

B

Explanation:
B. A type of artificial intelligence that can create new content and ideas, including text, images, music, and code. Generative AI refers to AI systems that can Generate original content Produce text, images, audio, video, and code Create outputs based on patterns learned from data

D

Explanation:
D is correct because it encapsulates the core functionality of an AI agent in generative AI, which is to autonomously analyze contexts and leverage tools for decision-making without needing continuous human intervention. Reasoning for D: AI agents function as autonomous decision-makers, processing varying inputs to evaluate situations and then employing learned strategies to arrive at conclusions. This self-sufficiency is vital in automation, allowing businesses to scale operations, optimize productivity, and reduce operational latency. Generative AI, specifically, benefits from this capacity as it enhances creativity and problem-solving by synthesizing new ideas or solutions based on pre-existing data and learned patterns. Evaluation of Other Options: A. To provide the computational resources needed to train and run gen AI models. This option describes the infrastructure necessary for AI operations but does not define the agent's role in decision-making or automation. Computational resources are a prerequisite rather than a defining function of an AI agent. B. To store and manage large datasets used for training and running gen AI models. Similar to option A, this focuses on data management rather than the agent's interactive or decision-making capabilities. While data management is crucial for training AI, it does not encapsulate the essence of what an AI agent does autonomously. C. To provide a user-friendly interface for interacting with gen AI models. While a user-friendly interface is important for human-AI interaction, it misrepresents the AI agent's primary ability to operate independently and make informed decisions. Such interfaces are supportive tools rather than core functions of an agent. In summary, D accurately defines the multifunctional role of an AI agent in generative AI by emphasizing its autonomy and decision-making capabilities, which are crucial for effective task automation. References: https://www.ijcai.org/Proceedings/2021/0422.pdf https://www.sciencedirect.com/science/article/pii/S2352864823006754 https://dl.acm.org/doi/10.1145/3488585.3488587

A

Explanation:
The correct answer is A, as it encapsulates the essence of an AI agent as an application designed to learn and achieve specific goals through available resources. AI agents operate within frameworks that highlight their ability to learn from interactions and adapt to varying contexts. This aligns with foundational principles in artificial intelligence, where systems are structured to evaluate input data continuously, refine their algorithms, and optimize their performance towards a defined objective. The definition inherently involves cognition-like capabilities, enabling agents to self-improve based on feedback loops, a principle evident in reinforcement learning. Critique of Other Options: B. A human expert who oversees and manages the operation of a gen AI application. This option mischaracterizes AI agents by attributing the role of agents to humans, thereby neglecting the autonomous operational capacity that defines AI entities. Human oversight is indeed valuable but does not alter the fundamental nature of an AI agent as an autonomous learner. C. A type of user interface that allows users to interact with a gen AI application. This description conflates user interface design with AI functionalities. While user interfaces facilitate interaction, they do not encompass the learning and goal-directed behaviors characteristic of AI agents, which operate beyond simple user interactions. D. A static component of a gen AI application that provides information to the user. This option fundamentally contradicts the dynamic nature of AI agents. By describing them as static components, it overlooks their ability to evolve their functionalities and adapt based on new information and experiences, which is essential for goal attainment. References: https://www.aaai.org/ojs/index.php/aimagazine/article/view/2752 https://medium.com/@atiik.k/what-is-an-ai-agent-29b5bcf1eaad https://www.sciencedirect.com/science/article/pii/S1568494623000364

B

Explanation:
B is correct because reinforcement learning (RL) is specifically designed for environments where agents learn from the consequences of their actions to maximize cumulative rewards. Reinforcement learning operates on the principles of trial and error, allowing AI agents to explore complex environments and make decisions based on received feedback (rewards or penalties). In this scenario, the AI character's ability to navigate and make decisions corresponds directly with RL paradigms, where the agent (AI character) learns a policy to maximize its score (reward) through iterative interaction with its environment. The positive and negative scores received during gameplay serve as the feedback mechanism necessary for RL algorithms to refine their strategies over time, promoting efficient learning to achieve specific objectives. Evaluation of Other Options: A. Unsupervised learning: This approach focuses on discovering patterns or structures in data without explicit labels. It does not utilize a feedback mechanism based on rewards or penalties, making it unsuitable for scenarios requiring decision-making and learning from interactions. C. Supervised learning: In supervised learning, algorithms learn from labeled datasets, requiring predefined input-output mappings. This method lacks the dynamic interaction with an environment, which is essential for developing adaptive strategies in gameplay, rendering it ineffective for real-time decision-making. D. Deep learning: While deep learning can be employed in RL contexts (specifically deep reinforcement learning), it is not an independent learning paradigm. Alone, it does not provide a framework for reward-based learning or trial-and-error experimentation characteristic of RL. For reinforcement learning applications: https://www.oreilly.com/library/view/reinforcement-learning/9781680991859/ https://www.tensorflow.org/tutorials/reinforcement_learning/actor_critic https://www.ibm.com/cloud/learn/reinforcement-learning

D

Explanation:
D is correct because unsupervised learning effectively identifies patterns or anomalies in unlabeled datasets. Unsupervised learning algorithms, such as clustering and dimensionality reduction, can analyze the inherent structure of the sensor readings without prior knowledge of labels. This is crucial in scenarios like the one described, where the dataset lacks predefined categories. Techniques like k-means clustering or principal component analysis can reveal natural groupings or outliers, helping the research team assess machine behavior and detect potential failures. In contrast, option A (Deep learning) is not suitable as it typically requires large amounts of labeled data for supervised tasks, and while it can work with unlabeled data, its complexity is unnecessary for initial anomaly detection in this scenario. B (Reinforcement learning) focuses on learning optimal actions based on reward feedback from an environment, which is irrelevant when the problem concerns data without actionable feedback or labels. Lastly, C (Supervised learning) necessitates a labeled dataset, making it infeasible for this context, where the team has yet to categorize the data. In summary, unsupervised learning is the most appropriate method for discovering patterns and anomalies within the given unlabeled sensor dataset. References: https://www.sas.com/en_us/insights/analytics/machine-learning.html https://towardsdatascience.com/a-beginners-guide-to-unsupervised-learning-29b0df3c6226 https://www.analyticsvidhya.com/blog/2021/06/unsupervised-learning-an-overview/