Foundation · 1 day
Generative AI Fundamentals
A technically accurate, practical introduction to modern generative AI for mixed technical and non-technical audiences. Enough depth to make sound decisions, with no requirement to write code.
1 day · Foundation · Foundations · Mixed technical and non-technical professionals
Overview
Most introductions to generative AI either stay at the level of product demonstrations or assume a software engineering background. This course sits between those two. It explains what large language models actually do, where they are reliable, and where they fail — in language that holds up in a technical conversation, without turning the day into a coding class.
You work through tokens, context, generation, prompting, retrieval, tools and agents as a coherent picture rather than a list of features. The aim is not familiarity with a particular product. It is a working understanding of the technology: how output is produced, why hallucination occurs, what retrieval actually changes, and what an agent means in practice.
The course is hands-on. You use current AI systems to test prompting, structured requests, retrieval and multimodal inputs, then examine the results with enough scepticism to be useful at work. Privacy, security and responsible use are treated as operating constraints, not as a closing presentation.
By the end of the day, you can talk about generative AI with precision, choose appropriate uses, and recognise designs that will not survive contact with real data, real users or organisational risk.
Audience
Mixed technical and non-technical professionals: product managers, analysts, knowledge workers, business stakeholders, and engineers who want a shared, accurate foundation before more specialised work.
Prerequisites
No programming experience is required. You should be comfortable using a computer and working with documents, and should have access to a current generative AI system during the course.
Duration
1 day
Can be adapted to a half-day briefing, or extended with additional exercises for a particular domain.
Learning outcomes
- 01
Explain how large language models generate output, including tokens, context and sampling, without relying on marketing language.
- 02
Diagnose common failure modes, including hallucination, stale knowledge and over-confident answers.
- 03
Design prompts and structured requests that produce more reliable, inspectable results.
- 04
Compare prompting, retrieval-augmented generation, tool use and agents, and say when each is justified.
- 05
Evaluate open and closed model choices against privacy, cost, capability and control.
- 06
Identify practical business uses that fit current capabilities, and reject uses that do not.
- 07
Apply responsible-use constraints around confidential information, verification and human accountability.
Outline
- 01
What generative AI is — and is not
- A precise picture of generative systems versus classical software and predictive ML.
- Large language models as next-token predictors, not knowledge bases or reasoning engines.
- Strengths: drafting, transformation, synthesis, coding assistance, interface generation.
- Limits: truth, currency, counting, long-horizon planning, and tasks that require authority.
- Exercise: inspect strong and weak outputs from the same task and name the failure mode.
- 02
How models produce output
- Tokens, context windows and why “the model read the document” is often the wrong description.
- Sampling, temperature and the difference between variation and correctness.
- Instruction following, system prompts and the gap between requested behaviour and guaranteed behaviour.
- Multimodal inputs: images, documents and the additional failure modes they introduce.
- Exercise: change context and sampling and observe how answers shift.
- 03
Prompting that can be inspected
- Role, task, constraints, examples and output shape as a practical prompting model.
- Structured prompting: lists, schemas, rubrics and checklists rather than open-ended chat.
- Iteration: testing a prompt against several inputs instead of polishing a single lucky answer.
- When better prompting is enough, and when the problem is missing information or missing tools.
- Exercise: rewrite a vague request into a structured prompt and compare the results.
- 04
Retrieval, tools and agents
- Why models cannot be assumed to know your documents, policies or systems.
- Retrieval-augmented generation: what is retrieved, what is generated, and what still has to be checked.
- Tools and function calling: giving a model access to search, calculators, tickets or APIs.
- Agents: loops over tools and state, including the difference between a useful workflow and an unsupervised one.
- Exercise: compare an ungrounded answer, a retrieved answer and a tool-using answer to the same question.
- 05
Models, risk and use in organisations
- Model families and the practical differences between hosted and self-hosted, open and closed.
- Privacy, data handling, retention and the difference between a consumer tool and an organisational deployment.
- Security and responsible use: confidential information, verification, provenance and human sign-off.
- Business use cases that hold up: drafting, knowledge access, analysis support, software assistance.
- Where the technology is heading, without treating research demos as production capability.
- Exercise: assess three proposed uses for value, risk and fitness to current systems.
Practical work
You work throughout the day in a current generative AI system. Exercises include inspecting failure modes, rewriting prompts, comparing ungrounded and retrieved answers, and assessing proposed workplace uses against capability, risk and data sensitivity.
Takeaways
- A shared, technically accurate vocabulary for models, context, retrieval, tools and agents
- Prompt patterns that can be reused and inspected at work
- A simple framework for judging whether a use case belongs to prompting, retrieval, tools or a conventional system
- A checklist for privacy, verification and responsible use
Delivery
Cognitecta delivers private corporate training, on-site or as remote live training. Courses can be run as published, or adapted to your organisation’s stack, domain and experience level.
Instructor
Nicholas Johnson, AI architect and software engineer. He has a degree in Artificial Intelligence and around twenty years of professional technology training, including hundreds of courses for engineering teams and large organisations. About.
Related courses
1 day · Foundation
Practical Generative AI at Work
A working day for knowledge workers and business teams: using generative AI for research, writing, analysis and document work, with enough discipline to trust the output.
1 day · Intermediate
How Large Language Models Work
A one-day explanation of LLM internals for developers, architects and technically curious professionals. Enough mechanism to improve engineering decisions, without a research-paper prerequisite.