A.I. – BRAHMAN

“The question of whether a computer can think is no more interesting than the question of whether a submarine can swim.” – Edsger Dijkstra

Deep Dive into LLMs like ChatGPT — Andrej Karpathy
Video · Foundations
An end-to-end explanation of how language models become chatbots. Watch in several sittings.
Transformers, the tech behind LLMs — 3Blue1Brown
Video + illustrated article · Foundations
Build visual intuition for tokens, embeddings, and transformers.
AI Evals: Everything You Need to Know — Hamel Husain
Article · Product quality
Learn how to identify failures, define good outputs, and measure whether an AI product is improving.
Building Effective Agents — Anthropic
Article · Agents
Understand workflows versus agents and decide how much autonomy a product needs.
People + AI Guidebook — Google PAIR
Guide · Product design
Explore user needs, expectations, feedback, and graceful handling of AI errors.

Attention in transformers, step-by-step — 3Blue1Brown
Video + illustrated article · Model fundamentals
Revisit attention through a detailed visual explanation before reading the original transformer paper.
The Illustrated Word2vec — Jay Alammar
Visual article · Embeddings
Understand how words become vectors: groundwork for semantic search and retrieval.
Effective Context Engineering for AI Agents — Anthropic
Article · Agents
Learn how to select and manage the information an agent needs to perform a task.
Contextual Retrieval — Anthropic
Article · RAG
Explore chunking, retrieval, and reranking. Ask whether a failure comes from finding information or using it.
Demystifying Evals for AI Agents — Anthropic
Article · Evaluation
Extend evaluation to systems that take multiple actions and use tools. Read after Hamel’s evals guide.

Attention is all you need – paper
Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities. – paper
A few useful things about Machine learning – paper
Dall-E – denoising diffusion – paper
Agentic patterns – link
Open AI Platform documentation – link
Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks
Paper · RAG
Core idea: combine a language model with retrieved information. PM question: how can our product use knowledge beyond its training data?
Training Language Models to Follow Instructions with Human Feedback
Paper · InstructGPT / RLHF
Core idea: use human feedback to improve instruction following. PM question: whose preferences define a good answer?
ReAct: Synergizing Reasoning and Acting in Language Models
Paper · Agents
Core idea: interleave reasoning with actions and observations. PM question: what changes when AI can use tools?

Taught by Afshine Amidi and Shervine Amidi · 9 lectures · Full playlist on YouTube · Course syllabus & slides

Lecture 1: Transformer (1:41:58)
Background on NLP and tasks, tokenization, embeddings, Word2vec/RNN/LSTM, attention mechanism, Transformer architecture
Lecture 2: Transformer-based models & tricks (1:47:19)
Attention approximation, MHA/MQA/GQA, position embeddings, RoPE, Transformer-based architectures, BERT and derivatives
Lecture 3: Large Language Models (1:48:44)
Definition and architecture, mixture of experts, context length and temperature, sampling, prompting and in-context learning, chain of thought, self-consistency
Lecture 4: LLM training (1:47:27)
Pretraining, quantization, hardware optimization, supervised finetuning (SFT), parameter-efficient finetuning (LoRA)
Lecture 5: LLM tuning (1:47:42)
Preference tuning, RLHF overview, reward modeling, PPO and variants, DPO
Lecture 6: LLM reasoning (1:47:10)
Reasoning models, RL for reasoning, GRPO, scaling
Lecture 7: Agentic LLMs (1:49:23)
Retrieval-augmented generation, advanced RAG, function calling, agents, ReAct framework
Lecture 8: LLM evaluation (1:49:25)
LLM-as-a-judge overview, best practices and benefits, biases and pitfalls
Lecture 9: Current trends (1:51:31)
Recap, trending topics, closing thoughts

Learning lists – Courses

Learning lists – Books

Artificial Intelligence: A Modern Approach, Stuart Russell and Peter Norvig – https://aima.cs.berkeley.edu
Understanding Deep Learning – https://udlbook.github.io/udlbook