Conceptual
Login

How Professionals Use AI: Text Generation

How professionals use AI text generation without being fooled, scoring a prompt against a dataset before trusting the output. You learn to specify the job instead of hunting for magic phrases, move from prompt engineering to context engineering, and match review effort to the cost of being wrong.

Estimated Time to Complete

Only available after login

Concepts

Retrieval-Augmented Generation (RAG) Tool-Augmented Language Models and Function Calling Matching Review Effort to the Cost of Being Wrong Source-Bound Generation: Restricting Output to a Supplied Corpus Conversational Agents and Dialogue Systems Where Prompt Tricks Still Pay: Small and Local Models Error Analysis: Reading Traces Until Failure Modes Stop Appearing Hallucination in Large Language Models In-Context Learning in Large Language Models Crafting Effective Prompts Large Language Models Predict Next Words Using Transformer Attention Mechanisms Context Layout and Ordering: Where in the Window Information Lands Improving a Prompt from Its Failure Cases Rather Than by Rewording Scoring a Prompt Against a Dataset Instead of Eyeballing One Output Task Specification: Job, Audience, Constraints, Output Shape Human Evaluation and Rank Correlation in NLP Prompt Engineering for Text Generation 6 of 47 Concepts are outdated Superseded by Context Engineering as the Successor to Prompt Engineering. The typed prompt is one of several context sources — system instructions, retrieved documents, tool results and history — and Role-Play Persona Preambles in Prompts 6 of 47 Concepts are outdated Superseded by Task Specification: Job, Audience, Constraints, Output Shape. Current vendor guidance calls heavy role prompting unnecessary and notes an over-specific persona can narrow rather than imp Hallucination Risk and Verification in AI Research Workflows Assembling a Small Labelled Eval Set Before a Prompt Reaches Production Machine-Checkable Output Contracts for Generated Text Worked Examples in the Prompt as a Specification of the Output You Want Temperature and Sampling in Language Model Decoding Sycophancy in Large Language Models Chain-of-Thought Prompting 6 of 47 Concepts are outdated Superseded by Controlling Reasoning Effort Instead of Scripting the Reasoning: frontier models reason internally, and vendor guidance now says typed 'think step by step' is redundant and can degrade r Chunking and Retrieval Quality as the Ceiling on a Grounded Answer NotebookLM as a Closed-Corpus Research Assistant Magic Phrases and Incentive Framing in Prompts 6 of 47 Concepts are outdated Superseded by Task Specification: Job, Audience, Constraints, Output Shape. Instruction tuning and RLHF absorbed most wording folklore, so stating the requirement plainly now outperforms incantation. Chain-of-Thought as a Model Behaviour You Can Read, Not a Phrase You Type LLM-Based Autonomous Agents Guarding Against Regression When a Prompt or Model Changes Uncertainty Estimation in Large Language Models Compaction: Summarising a Long Session to Reclaim Context Self-Refinement in Large Language Models Controlling Reasoning Effort Instead of Scripting the Reasoning Corpus Poisoning and Prompt-Injection Attacks on Retrieval Systems The Context Window as a Finite Budget You Spend Token-Efficient Prompt Engineering Strategies 6 of 47 Concepts are outdated Superseded by Context Engineering as the Successor to Prompt Engineering. Caching and cheap long windows moved the cost curve; the binding constraint is now context rot and retrieval precision, not pr Context Engineering as the Successor to Prompt Engineering Context Rot: Accuracy Decay Well Before the Window Is Full Long-Context Language Modeling Confidence Calibration in Classification Models Checking That a Citation Supports the Sentence Attached to It Automatic Evaluation Metrics for Natural Language Generation 6 of 47 Concepts are outdated Superseded by LLM-as-Judge Grading Calibrated Against Human Labels: BLEU-style n-gram overlap tracks human judgement poorly on open-ended generation, so professionals now use calibrated judges. Model Context Protocol as a Standard Way to Hand a Model Tools LLM-as-Judge Grading Calibrated Against Human Labels Automatic Evaluation and Benchmarking of Large Language Models