Fundamentals of gen AI · Topic 1 of 6
Core Gen AI Concepts
Generative AI Leader definitions: AI, NLP, machine learning, gen AI, foundation and multimodal models, LLMs, diffusion models, prompt engineering and tuning.
- 4 min
- AI is the field; ML learns from data; gen AI creates content.
- NLP is about human language.
- Foundation models are large, broadly trained and adaptable; LLMs focus on text; multimodal models handle several data types.
What it is
Generative AI sits inside a family of related ideas. The exam expects you to define each one in plain words: artificial intelligence, natural language processing, machine learning, generative AI, foundation models, multimodal foundation models, large language models, diffusion models, prompt engineering and prompt tuning.
How it works
The definitions
| Term | Plain meaning | Example |
|---|---|---|
| Artificial intelligence (AI) | The broad field of building systems that perform tasks needing human-like intelligence | Recognizing faces, understanding speech |
| Natural language processing (NLP) | The part of AI that lets computers understand and produce human language | Sentiment analysis of reviews, translation |
| Machine learning (ML) | Systems that learn patterns from data instead of following hand-written rules | Predicting customer churn |
| Generative AI (gen AI) | AI that creates new content: text, images, code, audio, video | Drafting a marketing email |
| Foundation model | A large model trained on broad data that can be adapted to many tasks | One model that summarizes, translates and answers questions |
| Large language model (LLM) | A foundation model specialized in understanding and generating text | A chat assistant answering policy questions |
| Multimodal foundation model | A foundation model that works across several types of input and output, such as text, images, audio and video | Describing what happens in a video, or answering questions about a chart |
| Diffusion model | A generative model, mostly for images and video, that learns to create content by turning random noise step by step into a clear result | Generating a product image from a text description |
Every LLM is a foundation model, but not every foundation model is an LLM. Image and multimodal models are foundation models too.
How foundation models change the economics
Before foundation models, each AI task needed its own model, its own labeled data and its own team. A foundation model is trained once, at great cost, by a provider such as Google, and many organizations then adapt it to their needs. That is why gen AI spread so fast: businesses can start from a capable model instead of from zero.
Prompt engineering and prompt tuning
| Prompt engineering | Prompt tuning | |
|---|---|---|
| What changes | The text of the prompt you write | A small set of learned values added to the model's input, trained on examples |
| Who does it | Anyone writing prompts | Developers, using a tuning job and example data |
| Model weights | Unchanged | Unchanged (only the small added values are learned) |
| Effort and cost | Lowest | Low compared with full fine-tuning |
| Example | Adding "Answer in three bullet points for a non-technical reader" | Teaching a model a company's classification style from a few hundred examples |
Prompt engineering is the craft of designing inputs that get the best output from a model. Prompt tuning is a parameter-efficient way of adapting a model: instead of retraining the whole model, it learns a small "soft prompt" from your examples. Both are cheaper than fine-tuning, which updates the model's own weights.
How the pieces relate
Gen AI is a branch of ML, which is a branch of AI. NLP overlaps with all of them: modern NLP is mostly done with LLMs. Foundation models are the engines of gen AI; LLMs, multimodal models and diffusion models are kinds of foundation model.
A worked scenario
A retailer wants to write product descriptions, create lifestyle images of its products and answer customer questions about photos they upload. Descriptions need an LLM; images need an image generation model, typically a diffusion model; answering questions about photos needs a multimodal foundation model. The marketing team improves results by prompt engineering first, then uses prompt tuning on approved descriptions to match the brand voice.
How the exam asks about this
- A capability is described and you name the term (diffusion model, multimodal model, NLP).
- "Which approach adapts a model without changing its weights or writing new prompts by hand?" Prompt tuning.
- "Why can a business adopt gen AI quickly?" Foundation models can be adapted instead of trained from scratch.
What to remember for the exam
- AI is the field; ML learns from data; gen AI creates content.
- NLP is about human language.
- Foundation models are large, broadly trained and adaptable; LLMs focus on text; multimodal models handle several data types.
- Diffusion models generate images and video by refining noise.
- Prompt engineering changes the prompt; prompt tuning learns a small soft prompt; fine-tuning changes model weights.
Do not confuse prompt engineering with prompt tuning. Engineering is writing better prompts; tuning is a training process on example data.
Common mistakes
Calling every gen AI model an LLM. Image and video generators are not language models.
Thinking a multimodal model just means many models glued together. One model handles several data types natively.
Assuming adaptation requires retraining the whole model. Prompting and prompt tuning adapt behavior cheaply.
Check yourself
What is a multimodal foundation model?
A foundation model that understands and generates several types of data, such as text, images, audio and video.
How does a diffusion model create an image?
It starts from random noise and refines it step by step into an image that matches the prompt.
What is the difference between prompt engineering and prompt tuning?
Prompt engineering writes better prompts; prompt tuning learns a small set of added values from example data without changing the model's weights.
Topic 1 of 6 in Fundamentals of gen AI
Test yourself on Fundamentals of gen AI
The full Google Cloud Generative AI Leader question bank has 512 questions, with timed practice exams and an explanation for every answer.
This study guide is independent and is not affiliated with, endorsed by, or sponsored by Google Cloud. Vendor names, certification names and exam codes are trademarks of their respective owners and are used for identification only.