LLM 101
A clear primer on large language models: their history from n-grams to the Transformer to ChatGPT, how they work (tokens, attention, context windows), the training pipeline (pretraining, SFT, RLHF, alignment), prompting and RAG, the key companies and people, and the numbers and milestones that define the field.
604 cards10 lessonsFree
Already have FlashFeed?Open this pack
Get this pack on your phone
Scan with your phone's camera.
flashfeedlearningapp.com/packs/llm-101
Part of the Learn AI collection.
10 lessons.
Each lesson has a study guide to read and objectives you can test yourself on. Its cards and games show up in your feed. Try one of its cards right here. Play a few of its games here too.
-
Unsplash · Glen Carrie · Unsplash License ↗01A Brief History of LLMs
Before the Transformer · The GPT era · ChatGPT and the modern wave
Lesson 1 · 10 cards1 / 10What did the Chinchilla (2022) result revise about GPT-3?
Chinchilla showed GPT-3 was undertrained given its 175B parameters: the 70B-parameter Chinchilla, trained on about four times as much data as the 280B Gopher at the same compute budget, outperformed both Gopher and GPT-3. Chinchilla revised the power laws rather than rejecting them.
Lesson 1 has 58 cards. Keep going in the free app.
Question -
FlashFeed original diagram · CC-BY-4.0
02How LLMs Work
Tokens, embeddings and attention · How text is generated
Lesson 2 · 5 cards1 / 5On average, roughly how much of a word does a single token represent?
A token is the basic unit an LLM processes — in English, on average about three-quarters of a word. Common words like 'the' are one token, while rarer strings like 'ChatGPT' split into several pieces. Tokens (not words) are what context limits, pricing, and speed are measured in.
Lesson 2 has 64 cards. Keep going in the free app.
Question -
AI-generated (gpt-image-1.5)03The Training Pipeline
Pretraining to RLHF · Alignment, inference and compute
Lesson 3 · 5 cards1 / 5Which is NOT true of the base model produced by pretraining?
Pretraining yields broad world knowledge and text prediction, but explicitly does NOT teach instruction-following or helpfulness — that requires SFT and RLHF. Base models are raw material, not safe for general users.
Lesson 3 has 54 cards. Keep going in the free app.
Question -
Pexels · Diana ✨ ↗04Prompting, RAG and Using LLMs
Prompting techniques · RAG, tools and agents
Lesson 4 · 5 cards1 / 5What is the key limitation of prompting alone?
Prompting costs nothing but is bounded by context and the model's existing weights, so it can't supply knowledge the model never learned. That gap is exactly why RAG, function calling, and agents exist.
Lesson 4 has 60 cards. Keep going in the free app.
Question -
05
Companies, People and Models
The companies · The godfathers and key researchers · Founders, leaders and the ecosystem
Lesson 5 · a questionWhich model family is made by xAI, the lab Elon Musk founded in 2023?
xAI makes Grok, which is integrated with X. Claude is Anthropic's, Llama is Meta AI's, and Gemini is Google DeepMind's.
Lesson 5 has 61 cards. Keep going in the free app.
- Take the lessons with you. Free, no account.Download on the App Store ↗Get it on Google Play ↗
-
06
Key Numbers and Milestones
Model sizes and dates · Scaling, context and misconceptions
Lesson 6 · a questionWhich sequence correctly orders these milestones in time?
The spine of the modern timeline: GPT-3 in 2020, ChatGPT on Nov 30, 2022, and GPT-4 on March 14, 2023 — the under-three-year stretch in which LLMs went from research artifact to mass technology.
Lesson 6 has 59 cards. Keep going in the free app.
-
Pexels · cottonbro studio · Pexels License ↗07What LLMs Are Good For
Language and writing tasks · Knowledge, search and code · Work and everyday tasks · High-stakes domains and choosing well
Lesson 7 · 10 cards1 / 10Which of these is NOT listed among the language and writing tasks LLMs handle well?
Simplification, rewriting/tone-shifting, and classification are all listed language strengths. Needing exact arithmetic without tools is named as a marker of poor LLM fit.
Lesson 7 has 87 cards. Keep going in the free app.
Question -
AI-generated (gpt-image-1.5)08Limits and Responsible Use
Reliability and staleness · Bias, privacy and your data · Security and societal risks · Limits, safeguards and deployment
Lesson 8 · 10 cards1 / 10At which three stages does bias enter an LLM?
Bias enters at data collection, pretraining, and fine-tuning — three distinct stages, each partly mitigated but rarely eliminated. Tokenization and decoding are mechanics, not bias-entry stages.
Lesson 8 has 68 cards. Keep going in the free app.
Question -
09
Quick Recall
Core terms
Lesson 9 · a questionA token is the unit an LLM actually processes. Roughly how much text does one token represent?
A token averages about ¾ of a word — it is the unit models process and the currency of context limits and pricing. Tokens are usually sub-word pieces, so "exactly one word" is wrong.
Lesson 9 has 45 cards. Keep going in the free app.
-
10
Key Concepts
Key People · Key Things
Lesson 10 · a questionWhich researcher is NOT one of the three who shared the 2018 Turing Award for deep learning?
The 2018 Turing Award went jointly to Bengio, Hinton, and LeCun. Andrew Ng is influential in AI as a Google Brain co-founder and educator, but he was not among the three recipients.
Lesson 10 has 48 cards. Keep going in the free app.
What you'll be able to do 47 objectives, from the pack's lessons
- Trace the pre-Transformer progression from RNNs/LSTMs through the attention mechanism to the 2017 Transformer.
- Contrast the Kaplan (2020) and Chinchilla (2022) scaling laws and what each revised about optimal model training.
- Explain how BERT's bidirectional (masked) pretraining differed from GPT and identify the GPT-1 to GPT-3 progression.
- Describe how RLHF/InstructGPT and the ChatGPT and GPT-4 launches turned capable models into useful products.
- Recognize the modern open-weight wave (LLaMA, Mistral) and the frontier assistants (Claude, Gemini, GPT-4).
- Explain how text becomes tokens and then embedding vectors a model can compute over.
- Describe the Transformer, self-attention, context window, and parameters that make up a model's internals.
- Describe next-token prediction and autoregressive generation as the core loop that produces text.
- Explain how logits and the temperature knob shape which next token is sampled.
- Explain why LLMs hallucinate and recognize emergent abilities that appear only at scale.
- Explain self-supervised pretraining and why the resulting base model is not yet a helpful assistant.
- Describe supervised fine-tuning as training on prompt/response pairs to make a model follow instructions.
- Explain how RLHF and its reward model align a model, and how RLAIF and Constitutional AI scale that alignment.
- Identify the instruct/chat model as the aligned product of the full base-plus-SFT-plus-RLHF pipeline.
- Distinguish training from inference and relate the FLOPs compute cost to why training a large model is costly but querying it is cheap.
- Describe what a prompt and a system prompt contain, and why prompting, though free, can't give a model knowledge it wasn't trained on.
- Distinguish zero-shot, few-shot and chain-of-thought prompting, explain in-context learning, and recognize step-by-step reasoning as emergent.
- Explain RAG and its supporting components — vector databases and chunking — for grounding a model in external knowledge.
- Describe function calling and LLM agents, and how an agent's planning loop differs from a single-prompt interaction.
- Contrast fine-tuning with RAG as ways to add knowledge or behavior, one baking into weights and the other into context.
- Match the US frontier labs (OpenAI, Anthropic, Meta, xAI) to their flagship model families.
- Identify the international and second-tier labs (Google DeepMind, Mistral, Cohere) and the lab that drove open-source LLMs.
- Recognize the three 'godfathers of deep learning' who shared the 2018 Turing Award.
- Identify pivotal LLM researchers — Vaswani, Sutskever, and Karpathy — and their contributions.
- Match today's lab leaders (Altman, the Amodeis, Hassabis) and the infrastructure players (Nvidia, Hugging Face).
- Recall the parameter counts of the GPT line (GPT-1, GPT-2, GPT-3) and GPT-3's training-data size.
- Recall the parameter counts of other landmark models (BERT-large, LLaMA's 7B-to-65B family, Mistral 7B and PaLM) and compare them with GPT-3.
- Place the key release years in order: BERT, GPT-2, GPT-3, LLaMA, ChatGPT, and GPT-4.
- Recall the ChatGPT and GPT-4 launch years and ChatGPT's record two-month climb to 100 million users.
- Recall the scaling-law figures (Kaplan year, Chinchilla's ~20 tokens/parameter) and context-window growth, and why bigger isn't always better.
- Identify LLM strengths across language tasks — summarization, translation, rewriting, editing, classification, and simplification.
- Explain LLM knowledge/search use cases (research, semantic search, extraction, RAG) and coding assistance (Copilot speedup, explanation, debugging, tests).
- Identify everyday work and personal tasks LLMs assist — support, drafting, meeting notes, data cleaning, tutoring, brainstorming — with their catches.
- Apply the copilot pattern and the cost-of-verification framework to decide how much automation a task warrants.
- Judge good vs poor LLM fit and the human-review guardrails required in healthcare, legal, and finance.
- Explain why LLMs hallucinate and what a knowledge cutoff means for stale, confidently-stated answers.
- Identify where LLM bias enters (data collection, pretraining, fine-tuning) and how it shows up, such as the AAVE accuracy gap.
- Describe what can happen to data pasted into a consumer LLM and the memorization/PII-extraction risk.
- Explain prompt injection and jailbreaks, automation bias, and why tool-enabled agents widen the attack surface.
- Recognize red-teaming and model cards, explain how non-determinism limits reproducibility, and why filters alone don't make a model safe for high stakes.
- Recall the core architecture terms — Transformer, token, self-attention, and context window — instantly.
- Recall the training terms — pretraining, RLHF, and Chinchilla's ~20-tokens-per-parameter rule.
- Recall the usage terms — RAG, inference, hallucination, few-shot, and emergence.
- Recognize the three Turing-Award 'godfathers of deep learning' — Bengio, Hinton, and LeCun.
- Identify key researchers behind modern LLMs — Sutskever and Radford at OpenAI, and educator Andrew Ng.
- Recognize foundational techniques and architectures — word2vec, LSTM, and the Transformer.
- Recognize landmark models — BERT, GPT-3, and ChatGPT — including GPT-3's 175-billion-parameter scale.
Study guide.
The same study guide as in the app, one page for each lesson, with its key facts and a self-test.
Built from 2 sources.
4 of its 606 cards cite a source. These are the sources it cites.
Hover over a photo, or tap its i, to see where it came from. All 6:
- Lesson 1: Unsplash · Glen Carrie · Unsplash License ↗
- Lesson 2: FlashFeed original diagram · CC-BY-4.0
- Lesson 3: AI-generated (gpt-image-1.5)
- Lesson 4: Pexels · Diana ✨ ↗
- Lesson 7: Pexels · cottonbro studio · Pexels License ↗
- Lesson 8: AI-generated (gpt-image-1.5)
- Cards
- 604
- Lessons
- 10
- Version
- 1.1.21
- Language
- English
The whole pack is one file. Read the pack file ↗
FlashFeed is on the App Store and Google Play.
FlashFeed is free on iPhone and Android, and needs no account.























