RNN — architecture & working
2 / 2 papers min 5 · max 8 marksExplain the architecture of RNN & how it works?
Explain RNN and its limitations.
🎯 Attachment 1: How to Pass — For Exam Tension
Minimum study · 60-mark & 30-mark formatsBreathe. You do not need every topic, and you do not need full marks. Passing lives around 27 marks out of 60 — or about 12–14 out of 30. The plan below starts with the six topics that appear in both papers, then covers each format's attempt order — so even with mistakes, you stay safely above the pass band.
Step 1 — the minimum set (learn these first, in this order)
| Learn | Where it pays | Marks on the table |
| Six repeaters | GenAI definition · GPT architecture · GPT variants · CNN · RNN · Ethics | Questions in both papers |
| 60-mark format | Q1 (15, pick three) + any three of Q2–Q6 (45) | 60 attempted; pass ~27, distinction 45 |
| 30-mark format | Q1 (10, pick five definitions) + any two of Q2–Q5 (20) | 30 attempted; pass ~12–14 (confirm from college) |
Step 2 — how to attempt in the hall
Attempt order, 60-mark paper: Q1 (25 min) → GPT architecture + CNN (30) → strongest remaining (30) → check (15).
For the 30-mark paper: Q1 definitions first (15 min), then the two strongest 10-mark questions (20 min each), check in the remaining time.
Format warning, not a guarantee. The Winter 2026 format is unconfirmed — if it returns to 60 marks, study the Dec template; if it stays 30, study May. Never mix the two papers' marks. Nothing here guarantees a pass — but it is the highest-coverage, lowest-effort path the evidence supports: the six repeat topics first, then your strongest remaining questions.
⚡ Attachment 2: Inside The Papers
Read this first — two formats, one course
Dec 2025: Q1 any three × 5 = 15, + any three of Q2–Q6 (15 each), Q6 itself any three × 5 = 15. Total 60, 2 hours.
May 2026: Q1 any five × 2 = 10, + any two of Q2–Q5 (10 each, 5 + 5). Total 30, 1 hour.
If Winter 2026 returns to the 60-mark ESE format, the Dec paper is the template. If it stays 30 marks, the May paper is. Do not study one format's marks as if they were the other's. The tracker below scales topic coverage to the 60-mark scale as an estimate only.
📋 Attachment 3: Model paper — practice only, not the real paper
Predicted shape · practice onlyGen AI has no slot map: the same topics reappear in different questions and different sizes. Nine topics appear in both papers — the six in the minimum set, plus LLM, pre-training and tools. Two practice versions below, one per format, built from PYQ questions only. Neither is the Winter 2026 paper.
| Version | Practice with | Marks |
| 60-mark (Dec template) | Q1: definition + GPT variants + data importance · Q2: CNN + pre-training/fine-tuning · Q4: challenges + GPT architecture · Q5: Chat-GPT + LLM types · Q6: ethics + domains + self-attention | 60 |
| 30-mark (May template) | Q1: five 2-mark definitions (GenAI, GPT/Chat-GPT, LLM, pre-training, LSTM or tokenization) · any two of: CNN + GPT architecture · ethics + GPT variants · Transformer + five tools · RNN + GPT architecture | 30 |
Module 1 — Introduction to Generative AI, Language Models & LLM Architectures
Rank 2 · 12 questionsThe foundations: what Generative AI is, how language models work, and the architectures — RNN, LSTM and Transformer — that everything else is built on.
LLM — Large Language Models
2 / 2 papers min 2 · max 8 marksWhat is LLM and Types of LLM explain?
Define LLM.
Generative AI — definition & scope
2 / 2 papers min 2 · max 5 marksDefine Generative AI. Explain its scope with suitable examples?
Define Generative AI.
Challenges in Generative AI
Dec only 8 marksWhat are the challenges in Generative AI?
Importance in five real-world domains
Dec only 5 marksExplain the importance of Generative AI in any five real-world domains.
Traditional vs modern language models
Dec only 5 marksDifferentiate between traditional language models and modern language models.
Transformer architecture
May only 5 marksExplain the architecture and working of Transformer models.
LSTM
May only 2 marksWhat is LSTM?
Tokenization in transformers
May only 2 marksWhat is tokenization in transformer model?
Module 2 — GPT, ChatGPT, Ethics & Generative AI Tools
Rank 1 · 15 questionsGPT is the centre of this paper: its architecture is asked three times, and variants, pre-training, ChatGPT, self-attention and the multi-head mechanism all hang off it. Ethics and the tools sit in this module too.
GPT architecture & working
2 / 2 papers min 5 · max 7 marksWrite a short note on the architecture and working of GPT models.
Explain architecture and working of GPT models.
Explain GPT architecture and working. (Same question as May Q2b — attempt only one.)
GPT variants
2 / 2 papers 5 marks both timesList any three GPT variants and mention one use case for each.
Explain different GPT variants and their applications.
Ethics in Generative AI
2 / 2 papers 5 marks both timesEthical concerns related to Generative AI.
Explain ethical issues in Generative AI.
Generative AI tools / applications
2 / 2 papers 5 marks both timesExplain "text-to-image" tools with an example.
Explain any five Generative AI tools with examples.
Pre-training
2 / 2 papers min 2 · max 7 marksExplain pre-training and fine tuning and instruct tuning in gen-AI?
What is pre-training?
Chat-GPT
Dec only 7 marksWhat is chat-GPT and how does it work?
Multi-head mechanism in GPT
Dec only 7 marksWhat is the multiread mechanism in GPT? (Printed as “multiread” — read as the multi-head attention mechanism.)
Self-attention in GPT
Dec only 5 marksExplain the importance and role of self-attention in GPT models.
GPT and Chat-GPT — definition
May only 2 marksWhat is GPT and Chat-GPT?
Asked anyway — not in the syllabus
Rank 3 · 3 questionsThese two topics are not named anywhere in the official syllabus — but both papers ask them anyway. Learn them regardless; do not skip them because they are unplaced.
CNN — architecture & working
2 / 2 papers min 5 · max 8 marksExplain How CNN Works?
Explain the architecture of CNN & how it works?
Importance of data in training
Dec only 5 marksWhat is the importance of data in training the AI model?
💡 Attachment 5: Step-by-Step Solutions
Solutions for Gen AI. Step-by-step solutions land here, one per question, in the same order as the ranked topics above. Clean, verified steps for every question.
Solution: Q1 — short skills
Step-by-step verified explanations in this order.
Solution: Module 1 — Introduction to Generative AI, Language Models & LLM Architectures
Step-by-step verified explanations in this order.
Solution: Module 2 — GPT, ChatGPT, Ethics & Generative AI Tools
Step-by-step verified explanations in this order.
Solution: Asked anyway — not in the syllabus
Step-by-step verified explanations in this order.
🌐 Attachment 6: All Subjects & 2-Tap Jumpgrid
Every subject follows the exact same 6 attachments: pass plan, cheat code, model paper, ranked PYQs, exact papers, and solutions.
| Subject | Ranked PYQs | Model paper | Pass plan | Exact papers | Solutions |
| Maths (N2113111) Computer · AI&DS · CS-IoT | Done — 42 questions | Done — MU pattern | Done — 30+ marks | Done — 2 papers | From tomorrow |
| DSGT (N2113112) Computer · AI&DS · CS-IoT | Done — 28 questions | Done — MU pattern | Done — 30+ marks | Done — 2 papers | From tomorrow |
| AOA (N2113113) Computer · AI&DS · CS-IoT | Done — 34 questions | Done — MU pattern | Done — 30+ marks | Done — 2 papers | From tomorrow |
| COA (N2113114) Computer · AI&DS · CS-IoT | Done — 29 questions | Done — MU pattern | Done — 30+ marks | Done — 2 papers | From tomorrow |
| Gen AI (OE, 2013311) Computer · AI&DS | Done — 30 questions | Done — 2 templates | Done — 27-45 marks | Done — 2 papers | From tomorrow |