Wrap-up and resources
Key Takeaways
- Responsible use first: any AI tool is as powerful as it serves your understanding, not how quickly it produces a final result.
- Clarity in the ask = accuracy in the answer: context, format, and breaking down complex tasks all reduce the model's room for guessing.
- Calibrated trust, not absolute: a language model is a very powerful statistical prediction tool for known patterns, and weak on precise rare facts or very recent knowledge, always verify based on the type of request.
- Retrieval builds memory: attempting the solution yourself, even with mistakes, cements understanding far better than reading a ready-made one.
- The model predicts, it does not calculate: generating one token at a time explains hallucination, math errors, the knowledge cutoff, and the size-versus-speed trade-off.
- Fluency is not evidence: Independent verification is a small habit that protects you from automation bias and the illusion of competence.
- Fix the prompt, not the output: Always ask: what did my prompt not specify? Then revise and repeat.
- Model ≠ platform: An answer may differ because of the model or the platform’s layers; compare with one prompt and record the trade-offs.
Glossary
| Term | Definition | Part |
|---|---|---|
| Hallucination | The model fabricating confident-sounding information that's actually incorrect. | 1 |
| Testing Effect / Retrieval Practice | A proven learning science finding: trying to retrieve information yourself cements it in long-term memory more effectively than rereading. | 1 |
| Prompt | The text you write to the model requesting a specific response. | 2 |
| Context | Background information you provide the model to guide its answer (who you are, what you know, what you want). | 2 |
| Zero-shot / Few-shot | Zero-shot: instructions with no examples. Few-shot: instructions with input/output examples for the model to follow the pattern. | 2 |
| Chain-of-Thought | A technique asking the model to show its reasoning steps one by one before the final answer. | 2 |
| Token | The smallest unit the model processes, can be a full word, a word fragment, or punctuation. | 3 |
| Parameter | A number inside the model used in computing each token, larger models have billions of parameters. | 3 |
| Context Window | The maximum number of tokens the model can see and process in a single conversation. | 3 |
| Inference | The phase where the trained model actually generates new text, token by token. | 3 |
| Temperature | A parameter controlling randomness in choosing the next token: zero = nearly deterministic, higher = more varied and creative. | 3 |
| Attention | A mathematical mechanism in the Transformer architecture letting each token distribute its "focus" across all other tokens in the text. | 3 |
| RLHF | Reinforcement Learning from Human Feedback: a training stage that improves the model through human preference rankings that train a reward model guiding outputs toward better answers. | 3 |
| Knowledge Cutoff | The date at which the model's training data stopped. | 3 |
| Trade-off | Balancing two conflicting advantages (like accuracy vs. speed) that can't both be maximized together. | 3 |
| Automation Bias | The tendency to prefer automated suggestions over one’s own observation. | 1 |
| Illusion of Competence | A feeling of mastery after smooth reading without testing understanding. | 1 |
| Role-based Prompting | Asking the model to take a role to change its style and assumptions. | 2 |
| Constraint | A limit in the prompt (audience, format, depth) that narrows the room for guessing. | 2 |
| Platform | A product wrapping a model with interface, safety, memory and tools. | 3 |
| VLM (Vision-Language Model) | A model that reads images with text and outputs text. | 3 |
| Quantization | Reducing the precision of model numbers to save memory at a small cost in quality. | 3 |
| Local model | A model running on your own machine without an outside service. | 3 |
Sources
The best outside resources for these lessons, classified by kind and level. All are open to everyone and need no account on any platform. Use the filters to narrow the list.
0 resources
Video (5)
VideoNeural networks (3Blue1Brown)A visual series on how networks learn, then how transformers work.3blue1brown.comBeginnerVideoBut what is a neural network? (3Blue1Brown)The best visual introduction to how a neural network recognises patterns.youtube.comBeginnerVideoLarge Language Models explained briefly (3Blue1Brown)In a few minutes: how a model predicts the next word and why that yields answers.youtube.comBeginnerVideo[1hr Talk] Intro to Large Language Models (Andrej Karpathy)What an LLM is, how it is trained, what it can and cannot do, and the security risks.youtube.comIntermediateVideoDeep Dive into LLMs like ChatGPT (Andrej Karpathy)The full pipeline from raw text to an assistant, including why models hallucinate.youtube.comIntermediate
Interactive tools (4)
Interactive toolThe tokenizer toolPaste text and see how it splits into tokens, numbers included.platform.openai.comBeginnerInteractive toolOllamaA hands-on tool that runs right in the browser.ollama.comIntermediateInteractive toolLMArenaA hands-on tool that runs right in the browser.lmarena.aiIntermediateInteractive toolPrompt Engineering Interactive Tutorial (Anthropic)Nine chapters with exercises: clear instructions, roles, examples and step-by-step reasoning.github.comBeginner
Official documentation (4)
Official docsPrompt engineering overviewHow to write a clear, specific request that gets a sharper answer.docs.anthropic.comBeginnerOfficial docsOECD, AI PrinciplesAn official reference covering the topic precisely.oecd.orgIntermediateOfficial docsOpenAI, Prompting guideAn official reference covering the topic precisely.platform.openai.comIntermediateOfficial docsGoogle Cloud, Prompt iteration strategiesAn official reference covering the topic precisely.docs.cloud.google.comIntermediate
Articles and explainers (18)
ArticleWhy language models hallucinateThe OpenAI research explaining hallucination: training effects and evaluations that reward guessing.openai.comIntermediateArticleThe testing effectWhy retrieving knowledge yourself fixes it better than rereading.en.wikipedia.orgBeginnerArticleThe assistant's guiding principlesA document laying out the values behind Claude's behaviour: honesty and avoiding harm.anthropic.comBeginnerArticleThe Prompt Engineering GuideAn open guide to prompting techniques, with examples.promptingguide.aiIntermediateArticleHow to ask a good questionOld, still-valid rules: summarise, include the error, show what you tried. They apply to the assistant too.stackoverflow.comBeginnerArticleThe Illustrated TransformerA drawn walkthrough of attention and the architecture behind these models.jalammar.github.ioIntermediateArticlesCSU AI Commons, Ethical and Responsible Use of AI for StudentsAn article explaining the topic with examples and clear steps.genai.calstate.eduIntermediateArticlesSuperblocks, What is Responsible AI? Ethical Policies, Practices & Top ToolsAn article explaining the topic with examples and clear steps.superblocks.comIntermediateArticlesPrompt Engineering Guide, General Tips for Designing PromptsAn article explaining the topic with examples and clear steps.promptingguide.aiIntermediateArticlesAnthropic, Prompt Engineering Best PracticesAn article explaining the topic with examples and clear steps.claude.comIntermediateArticlesDigitalOcean, Prompt Engineering Best Practices: Tips, Tricks, and ToolsAn article explaining the topic with examples and clear steps.digitalocean.comIntermediateArticlesCodeSignal, Prompt Engineering Best Practices 2025An article explaining the topic with examples and clear steps.codesignal.comIntermediateArticlesProjectPro, 10 Biggest Limitations of Large Language ModelsAn article explaining the topic with examples and clear steps.projectpro.ioIntermediateArticlesOpenReplay, How LLMs WorkAn article explaining the topic with examples and clear steps.blog.openreplay.comIntermediateArticlesAutomation bias, WikipediaAn encyclopedia article on the concept, terms and background.en.wikipedia.orgIntermediateArticlesStephen Wolfram, What Is ChatGPT Doing … and Why Does It Work?An article explaining the topic with examples and clear steps.writings.stephenwolfram.comAdvancedArticlesLaunchDarkly, Prompt Engineering Best PracticesAn article explaining the topic with examples and clear steps.launchdarkly.comIntermediateArticlesGeeksforGeeks, Role-based promptingA simple walkthrough with code examples, step by step.geeksforgeeks.orgBeginner
Resources for this lesson
Grouped by type. Start with the official docs for precision, or an article or video for the explanation.
Articles3
Done with this lesson?
Sign in to save your progress.