Myths, Glossary and Cheat Sheet
Keep this for later.
This final lesson is one to bookmark. We will clear up common myths, review the key words in plain language, and finish with a cheat sheet that sums up the whole course. Come back to it whenever you need a quick reminder.
- Common myths, cleared up
- Key words in plain language
- A cheat sheet to keep
Myth: AI understands like a person
Myth number one. A I thinks, and understands, exactly like a person. Reality, a language model predicts likely tokens, using patterns learned from data. It can behave as if it understands, and often does so impressively, but it has no body, no life experience, and no guaranteed grasp of truth. Treat its fluency as a strength, and a risk.
The myth
- It understands like a person
- It always knows when it is right
- Its confidence means accuracy
The reality
- It predicts likely tokens
- It can be fluent and wrong
- Verify anything important
Myth: bigger is always better
Myth number two. The biggest model is always the best choice. Reality, bigger models are stronger on hard tasks, but slower, and costlier. For simple, repeated jobs, a small model is often plenty, and far cheaper. And a smaller model, with good data, and good prompts, can beat a larger one, on your specific task. Choose by the job, not by the headline.
The myth
- Always use the biggest model
- Size equals quality for every task
The reality
- Match the model to the task
- Small and cheap often wins
- Test on your own examples
Myth: AI will be perfect soon
Myth number three. Soon, A I will never make mistakes. Reality, errors are built into how these models work, and they get rarer, but they do not vanish. That is why good systems add checks, sources, tests, and human review. The goal is not perfection. The goal is a reliable system, with sensible safeguards.
The myth
- Mistakes will disappear
- No need to check the answers
The reality
- Errors get rarer, not zero
- Checks, sources and tests matter
- Humans review what is important
Glossary: part one
Now, a glossary. A token, a small piece of text, the unit models read. An embedding, a list of numbers that captures meaning. Attention, how a model decides which words matter. A transformer, the design built on attention. Pre-training, learning to predict the next token, on huge data. And fine tuning, extra training, to adapt a model.
| Word | Plain meaning |
|---|---|
| Token | a small piece of text |
| Embedding | meaning as a list of numbers |
| Attention | which words matter to which |
| Transformer | the design built on attention |
| Pre-training | learning from huge data |
| Fine-tuning | extra training to adapt a model |
Glossary: part two
More words. A prompt, the instruction you give. The context window, the model's working memory. Temperature, the creativity dial. A hallucination, a confident, false answer. R A G, retrieving documents, before answering. An agent, a model that plans, uses tools, and loops. And alignment, making models helpful, honest, and harmless.
| Word | Plain meaning |
|---|---|
| Prompt | the instruction you give |
| Context window | the working memory |
| Temperature | the creativity dial |
| Hallucination | a confident, false answer |
| RAG | look it up, then answer |
| Agent | plans, uses tools, loops |
| Alignment | helpful, honest, harmless |
The whole course on one page
And here is the whole course on one page. A I learns patterns from data. Language models predict the next token, using transformers, and attention. Training turns raw text into an assistant, through pre training, instruction tuning, and human feedback. You steer it with prompts. You ground it with retrieval. You extend it with tools, and agents. You create images with diffusion. And you use it all responsibly, with checks, privacy, and care.
- Learnpatterns from data, by guess, check and adjust
- Predictthe next token, with transformers and attention
- Trainpre-train, instruction-tune, align
- Useprompts, retrieval, tools and agents
- Createimages, audio and video with diffusion
- Be responsiblecheck, protect, and stay accountable
Summary
That brings us to the end of this course. You started by wondering how a machine can write, and paint, and speak. Now you understand the ideas, tokens, embeddings, attention, training, and diffusion, and you know how to prompt, build, and use A I responsibly. Thank you for learning with me. Go and create something wonderful. Goodbye, and good luck.
Key takeaways- You know how it works
- You can prompt, build and customise
- Use it responsibly, and keep creating