GK Questions

Responsible MLOps: model cards document limitations, fairness audits detect bias across groups, A/B testing validates real-world impact. Critical for ethical AI deployment.

Model pruning removes low-importance weights/neurons, reducing size and inference cost. Combined with quantization for efficient edge deployment. Critical for scalable AI systems.

Prompt injection exploits LLM instruction-following to bypass safety guards, extract training data, or execute unauthorized actions. Critical vulnerability for LLM applications requiring robust input sanitization.

Reinforcement Learning from Human Feedback (RLHF) trains reward models from human rankings, then optimizes LLM to maximize rewards. Critical for aligning AI with human values and safety.

GPAI (founded 2020) brings together experts for responsible AI: research, policy, best practices. India is founding member. Critical for understanding global AI governance landscape.

Effective tech policy: technology-neutral (focus on outcomes), proportionate to risks, adaptable to innovation. Balances consumer protection with innovation incentives. Critical for policy design.

Digital diplomacy: cyber norms, data governance, tech partnerships, capacity building. India actively engages in UN, GPAI, OECD on AI ethics, cybersecurity, digital trade. Critical for foreign policy questions.

Telecommunications Standards Development Society, India (TSDSI) develops indigenous telecom standards, represents India in 3GPP, ITU. Critical for technology sovereignty and 5G/6G leadership.

India Stack layers: identity (Aadhaar), payments (UPI), data sharing (AA), documents (DigiLocker). Enables innovation on public digital infrastructure. Critical for digital governance questions.

Multiple imputation creates several completed datasets with different plausible values for missing data, analyzing each and combining results. Accounts for uncertainty vs single imputation methods.

Responsible data science: fairness audits for bias, privacy-preserving techniques, impact assessments. Critical for ethical AI deployment and regulatory compliance (DPDP Act).

Principal Component Analysis (PCA) transforms features into orthogonal components ordered by variance explained. Enables visualization, noise reduction, and faster model training.

Precision = TP / (TP + FP). Critical for imbalanced classification tasks where false positives are costly (fraud detection, medical diagnosis). Complements recall for comprehensive evaluation.

Histograms bin continuous data to show frequency distribution, revealing skewness, outliers, and modality. Critical for exploratory data analysis and statistical understanding.

IaC (Terraform, CloudFormation) defines infrastructure in declarative code, enabling version control, peer review, automated testing, and consistent deployments. Critical for cloud governance.

Refactor (re-architect) redesigns applications for cloud-native: microservices, serverless, containers. Maximizes cloud benefits but requires significant effort. Critical for digital transformation.

Hybrid cloud combines on-premises infrastructure with public cloud services. Enables data sovereignty compliance while leveraging cloud scalability. Critical for regulated industries.

Multi-cloud architecture distributes workloads across AWS, Azure, GCP to avoid vendor lock-in, improve resilience, and leverage best-of-breed services. Critical for enterprise cloud strategy.

SOAR (Security Orchestration, Automation, and Response) automates playbooks: alert triage, enrichment, containment actions. Reduces analyst workload and response time. Critical for SOC efficiency.

NIST SP 800-61 and ISO 27035 provide structured incident response methodologies. Critical for consistent, effective handling of security incidents in enterprises.