Chapter 7: Machine Learning and Artificial Intelligence · Term 2
Lesson 7-3 explained: Large Language Models (LLM) and Generative AI
A language model generates text by predicting the next word (plausible, not always true → hallucination). LLMs use vast data and billions to trillions of parameters; RLHF tunes them by human feedback; multimodality unifies text, image, and audio. Fluency ≠ truth — always verify.
Lesson summary
A language model generates text by predicting the next word again and again — choosing what is statistically plausible, not guaranteed true (the cause of hallucination). Large language models scale this up with vast data and billions to trillions of parameters, are tuned by RLHF to be helpful, and, through multimodality, handle text, images, and audio together.
Takeaway: A language model generates text by predicting the next word (plausible, not always true → hallucination). LLMs use vast data and billions to trillions of parameters; RLHF tunes them by human feedback; multimodality unifies text, image, and audio. Fluency ≠ truth — always verify.
Key terms
- Language model
- Hallucination
- Large language model (LLM)
- Parameter
- RLHF
- Multimodality
- Modality
- Generative AI
- Fact-check
Textbook page: Arabic book 67 · English book 77
Free questions on this lesson
From the textbook question bank. Think of the answer first, then open it.
1. What is a language model?
- A dictionary that only stores word meanings
- A table of ready-made sentence templates
- A model that learns from lots of text and predicts which word (token) is likely to come next in a sentence
- A program that turns images into sounds
Answer
Correct answer: A model that learns from lots of text and predicts which word (token) is likely to come next in a sentence
A language model learns word-usage patterns from text, then predicts what comes next given the context.
1. Language Models and "Predicting the Next Word" · page 78
2. Does a language model's large size guarantee its information is correct?
- No: it may still get facts wrong even if it succeeds at many tasks
- Yes, the larger the model the more error-free
- Yes, once it passes a billion parameters
- Yes, because its data comes from the Internet
Answer
Correct answer: No: it may still get facts wrong even if it succeeds at many tasks
Size may improve performance but does not guarantee truth, so checking stays necessary.
2. Abilities as scale grows · page 79
3. Where is generative AI heading thanks to multimodality?
- To a technology for numbers only
- From a language-only technology to one linking many kinds of information
- To needing no data
- To stopping text generation
Answer
Correct answer: From a language-only technology to one linking many kinds of information
Multimodality widens generative AI's uses to link text with images, audio and more.
4. How multimodality is achieved · page 81
Keep studying on the platform
The full explanation of every lesson, interactive simulations, homework and an exam per chapter, and the whole question bank. Try it free, no code needed.
Lessons in this chapter 7: Machine Learning and Artificial Intelligence
- 7-1 The Basics of Machine Learning
- 7-2 Neural Networks and Deep Learning
- 7-3 Large Language Models (LLM) and Generative AI