Margdarshak · The Guide
Where do you want
to go?
Not a firehose — a curriculum. Each track is a guided sequence of papers that builds understanding from first principles, with a reason for every paper in every position.
Start here
Essential
Go deeper
🧠
Intermediate5 papers· ~50 min
The Reasoning Revolution
Understand why o1 and DeepSeek-R1 represent a fundamentally new paradigm, and explain the research lineage that made them possible.
1DualCoT-VLA: Visual-Linguistic Chain of Thought via Parallel Reasoning for Vision-Language-Action Models
→2Learning When to Sample: Confidence-Aware Self-Consistency for Efficient LLM Chain-of-Thought Reasoning
→3Tree of Thoughts: Deliberate Problem Solving with Large Language Models
🤖
Intermediate5 papers· ~50 min
Build AI Agents
Understand the architecture behind any AI agent — how it reasons, uses tools, self-corrects, and operates in the real world.
1ReAct: Synergizing Reasoning and Acting in Language Models
→2Toolformer: Language Models Can Teach Themselves to Use Tools
→3Reflexion: Language Agents with Verbal Reinforcement Learning
Applied
⚡
Applied4 papers· ~40 min
Ship AI Without the GPU Bill
Understand why Mistral can compete with GPT-4 at a fraction of the cost, and evaluate any AI efficiency claim your engineers make.
1LoRA: Low-Rank Adaptation of Large Language Models
→2FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness
→3Mixtral of Experts
🛡️
Applied4 papers· ~40 min
AI Safety for Product Teams
Evaluate any AI system's safety properties, explain alignment to stakeholders, and make informed decisions about what to ship.
1Training language models to follow instructions with human feedback
→2Constitutional AI: Harmlessness from AI Feedback
→3Direct Preference Optimization: Your Language Model is Secretly a Reward Model