A quiz on the world of Artificial Intelligence

Comprehensive Prelims + Mains Study Notes

1. At a Glance

• Artificial Intelligence (AI) refers to the simulation of human cognitive functions — reasoning, learning, problem-solving, and language understanding — by machines, especially computer systems.
• The transformer architecture (2017) is the foundational breakthrough underpinning virtually all modern Large Language Models (LLMs) — ChatGPT, Gemini, Llama, Copilot — making AI a defining geopolitical and economic variable. [1]
• India's IndiaAI Mission (2024) with a ₹10,371.92 crore outlay positions AI as a GS-III and GS-II topic simultaneously — industrial policy, governance, ethics, and sovereignty all converge. [2]
• UPSC has tested AI concepts across Prelims (science-tech) and Mains (GS-III: technology; GS-IV: ethics of AI).

2. Why in the News

• The Hindu quiz (February 20, 2026) featured a dedicated AI quiz testing concepts like transformer architecture, NVIDIA Tensor Cores, Microsoft Copilot's Prometheus model, AlexNet, RNNs, LSTMs, and AI hallucination — signalling the mainstreaming of AI literacy in competitive discourse. [4]
• IndiaAI Mission, approved March 2024, crossed 38,000 GPUs (against initial target of 10,000), reflecting India's accelerated push into AI infrastructure. [2]
• India AI Governance Guidelines published for public consultation in early 2025 by MeitY's AI Governance Subcommittee (set up November 2023). [2]
• Global context: ChatGPT surpassed 100 million users in 2023; DeepSeek (China) emerged as a challenger to OpenAI in 2025-26, referenced in the quiz's visual question about an IITM alumnus whose firm challenges Google Search and ChatGPT. [4]

3. Background & Evolution

4. Core Static Facts

A. Key Terminology

• LLM (Large Language Model): Deep learning model trained on vast text corpora to generate and understand human language (e.g., GPT-4, Gemini, Claude).
• Transformer Architecture: Neural network design based purely on self-attention mechanisms, introduced in the 2017 paper "Attention Is All You Need" by Vaswani et al. (Google). Eliminates recurrence (RNNs) entirely. [1]
• Self-Attention: Mechanism allowing a model to weigh the importance of different words in a sequence globally — enabling parallel processing.
• RNN (Recurrent Neural Network): Neural network processing sequential data step-by-step; predecessor to transformers; used in NLP before 2017. [4]
• LSTM (Long Short-Term Memory): Variant of RNN designed to remember long-range dependencies; mitigates the "vanishing gradient" problem. [4]
• CNN (Convolutional Neural Network): Used in image recognition; basis of AlexNet (2012). [4]
• Hallucination (AI): When an AI model generates factually incorrect, fabricated, or nonsensical information presented as fact, due to pattern completion without grounding in truth. [4]
• RAG (Retrieval-Augmented Generation): Technique that grounds LLM responses in real-time retrieved documents (e.g., Copilot using Bing index). [4]
• Generative AI: Subset of AI that creates new content (text, images, audio, code) — distinct from discriminative AI.

B. Key Hardware

• GPU (Graphics Processing Unit): Massively parallel processor repurposed for AI/deep learning workloads.
• Tensor Cores (NVIDIA): Specialised processing units within NVIDIA GPUs designed to drastically accelerate AI, deep learning, and HPC workloads. [4]
• TPU (Tensor Processing Unit): Google's custom ASIC designed specifically for neural network inference and training.

C. Key Models / Products

D. India's AI Governance Architecture

• Nodal Ministry: MeitY (Ministry of Electronics and Information Technology) [2][3]
• IndiaAI Mission: Cabinet approval — March 2024; outlay ₹10,371.92 crore over 5 years [2]
• Seven Pillars of IndiaAI: (1) AI Compute, (2) Data for AI, (3) AI Innovation Centre, (4) AI Skilling, (5) AI Startup Financing, (6) Safe & Trusted AI, (7) AI for Governance [2][3]
• National Centre for AI: Key institutional anchor under IndiaAI Mission [3]
• AI Governance Guidelines Subcommittee: Set up November 2023 by MeitY; report published 2025 [2]
• Guiding Principles (AI Governance): Transparency, accountability, fairness, safety, inclusive innovation [2]
• India's AI Vision Statement: "Making AI in India and Making AI Work for India" [2]

5. Multi-Dimensional Analysis

Economic

• IndiaAI Mission targets building sovereign AI compute capacity — 38,000 GPUs achieved vs. 10,000 initial target — reducing dependence on US cloud providers. [2]
• AI projected to add $1 trillion to India's GDP by 2035 (NASSCOM/NITI Aayog estimates); labour displacement vs. job creation tension is central.
• NVIDIA's Tensor Core dominance (>80% AI chip market share) creates supply chain vulnerability for nations pursuing AI sovereignty. [4]

Geopolitical / Strategic

• US–China AI rivalry is the defining strategic contest: US controls frontier model training (OpenAI, Anthropic, Google); China has DeepSeek, Baidu Ernie.
• India's IndiaAI Mission explicitly targets AI sovereignty — domestic compute, domestic datasets, domestic models.
• Wassenaar Arrangement and US export controls on advanced semiconductors (H100/A100 GPUs) directly constrain India's ability to scale AI infrastructure.
• Microsoft's Prometheus (Copilot) and Bing integration represent the US big-tech model of AI-search convergence — reshaping information geopolitics. [4]

Legal / Constitutional

• India has no dedicated AI Act as of 2025 (unlike EU AI Act, 2024).
• Digital Personal Data Protection Act, 2023 (DPDPA) is the nearest statutory framework governing AI-processed personal data.
• IT Act, 2000 and IT (Intermediary Guidelines) Rules, 2021 have limited applicability to generative AI outputs.
• MeitY's AI Governance Guidelines (2025) are advisory/non-binding — no statutory force yet. [2]

Scientific / Technological

• Transformer architecture (2017, Google, "Attention Is All You Need") displaced RNNs/LSTMs by enabling parallelism and capturing long-range dependencies via self-attention. [1]
• AlexNet (2012): First deep CNN to win ImageNet; co-invented by Ilya Sutskever (Israeli-Canadian), Geoffrey Hinton, Alex Krizhevsky — directly enabled the modern deep learning era. [4]
• Hallucination remains an unsolved alignment problem: LLMs generate plausible-but-false outputs, posing risks in legal, medical, and governance domains. [4]
• Multimodal AI (GPT-4o, Gemini) processes text + image + audio simultaneously — next frontier beyond LLMs.

Ethical / Governance

• AI hallucination raises questions of liability — who is responsible when an AI gives wrong medical/legal advice? [4]
• Algorithmic bias: Models trained on skewed historical data replicate discrimination (caste, gender, race).
• India's AI Governance Guidelines stress transparency and accountability — aligned with UNESCO's 2021 Recommendation on the Ethics of AI. [2]
• Deepfakes (AI-generated synthetic media) threaten electoral integrity — SEBI, ECI have flagged concerns.

Administrative

• MeitY is the nodal ministry; IndiaAI is the implementation body; NASSCOM is an industry partner. [2][3]
• Federal dimension: AI policy is a Union subject; states lack independent AI governance frameworks.
• Digital Divide risk: IndiaAI Mission's skilling pillar must address rural/vernacular access gaps.

6. Recent Developments (last 12–18 months)

• November 2025: India AI Governance Guidelines published for public consultation by MeitY Subcommittee. [2]
• 2025–26: IndiaAI compute infrastructure reaches 38,000 GPUs — far exceeding initial 10,000 GPU target. [2]
• February 2026: The Hindu features AI quiz testing transformer architecture, Tensor Cores, Prometheus, AlexNet, RNN/LSTM, hallucination — signalling AI as mainstream competitive exam topic. [4]
• 2025: DeepSeek (China) disrupts AI landscape with open-weight model competitive with GPT-4 at fraction of cost — geopolitical and economic shock for US AI firms.
• 2024 (March): Cabinet approves IndiaAI Mission — ₹10,371.92 crore, 5-year horizon. [2]
• 2024: EU passes world's first comprehensive AI Act — risk-based regulatory framework; India watches closely as template.
• 2025: Ilya Sutskever (co-founder/former chief scientist, OpenAI; co-inventor of AlexNet) founds Safe Superintelligence Inc. (SSI) focused on AI safety research. [4]

7. Prelims Hooks (high-density factual bullets)

1. The seminal paper introducing the Transformer architecture is titled "Attention Is All You Need", published in 2017 by Google researchers (Vaswani et al.). [1]
2. The paper "Attention Is All You Need" was published originally on arXiv as preprint 1706.03762. [1]
3. Tensor Cores are specialised processing units within NVIDIA GPUs that accelerate AI, deep learning, and HPC workloads. [4]
4. Microsoft Copilot uses a proprietary orchestrator named Prometheus to connect its LLM to the Bing search index for real-time grounding. [4]
5. Ilya Sutskever — Israeli-Canadian scientist, co-founder and former chief scientist of OpenAI — co-invented AlexNet, a convolutional neural network (2012). [4]
6. RNN = Recurrent Neural Network; LSTM = Long Short-Term Memory — both used for sequential data and NLP tasks; LSTMs address the vanishing gradient problem of vanilla RNNs. [4]
7. AI Hallucination: An AI generating factually incorrect or fabricated content confidently presented as true. [4]
8. IndiaAI Mission approved March 2024; budget outlay ₹10,371.92 crore over 5 years; nodal ministry: MeitY. [2]
9. India's AI vision: "Making AI in India and Making AI Work for India" — IndiaAI Mission. [2]
10. IndiaAI compute infrastructure achieved 38,000 GPUs against initial target of 10,000 GPUs. [2]
11. MeitY's AI Governance Subcommittee was set up in November 2023; guidelines released for public consultation in 2025. [2]
12. AlexNet (2012) won the ImageNet challenge and is considered a watershed moment launching the deep learning era.
13. Transformers replaced RNNs by enabling parallel processing of entire sequences via self-attention, unlike RNNs' sequential step-by-step processing. [1]
14. India's Digital Personal Data Protection Act, 2023 (DPDPA) is the nearest statutory framework applicable to AI-processed personal data — India has no dedicated AI Act as of 2026. [2]

8. Mains Relevance

GS Paper Mapping:

Plausible Mains Question Stems:

1. "The rise of transformer-based Large Language Models (LLMs) presents both transformative opportunities and governance challenges for India. Critically analyse India's preparedness through the lens of the IndiaAI Mission and existing regulatory frameworks." (GS-III / 250 words)

2. "AI hallucination is not merely a technical bug but a governance crisis. Examine the ethical and legal implications of AI-generated misinformation and suggest a regulatory framework suitable for India." (GS-IV / 150 words)

3. "India's IndiaAI Mission seeks AI sovereignty through compute, data, and skilling. Evaluate its progress and identify structural bottlenecks in achieving the stated vision of 'Making AI Work for India'." (GS-II + GS-III / 250 words)

9. Related Topics to Study Next

10. Common Errors / Trap Areas

1. "Attention is All You Need" confused with a product: It is a research paper (2017), not a software or model. The architecture it introduced — the Transformer — is the foundation of GPT, BERT, Gemini, etc.

2. Ilya Sutskever vs. Geoffrey Hinton: Both are connected to deep learning and AlexNet. Ilya Sutskever is the Israeli-Canadian scientist who co-invented AlexNet and co-founded OpenAI. Geoffrey Hinton (Turing Award winner, "Godfather of AI") was Sutskever's supervisor but is British-Canadian, not Israeli-Canadian. Do not conflate them.

3. MeitY vs. NITI Aayog as nodal body: MeitY is the implementing ministry for IndiaAI Mission and AI Governance. NITI Aayog published the National Strategy for AI (2018) but is not the implementing body for IndiaAI Mission.

4. RNN vs. LSTM vs. Transformer: RNN is the broad class; LSTM is a variant of RNN (not a separate class); Transformer replaced both for most NLP tasks — it is not a type of RNN.

5. Hallucination ≠ Bias: AI hallucination = generating false facts confidently. AI bias = systematic discrimination from skewed training data. These are distinct failure modes with different regulatory responses — examiners test this distinction.