Generative AI is a type of artificial intelligence that creates new content, such as text, images, music, code, or video. It matters because many students now use tools that can answer questions, summarize ideas, make art, or help write computer programs. These systems do not think like humans, but they can find patterns in huge collections of examples and use those patterns to make something new.
Understanding how generative AI works helps students use it wisely and check its results carefully.
A generative AI model learns during training by adjusting many internal numbers called parameters so its predictions get better. For a text model, the system often predicts the next token, which may be a word, part of a word, or symbol, based on the tokens that came before it. When a user gives a prompt, the model uses probabilities to choose likely next tokens and builds an output step by step.
The result can be useful and creative, but it can also contain errors because the model is predicting patterns rather than verifying truth.
Key Facts
- Generative AI creates new outputs from learned patterns in training data.
- A prompt is the input that guides what the AI generates.
- For text generation, the model estimates P(next token | context).
- Training reduces error by updating parameters: new parameter = old parameter - learning rate x gradient.
- More training data can improve pattern learning, but data quality and fairness also matter.
- Generative AI output should be checked because confident answers can still be incorrect.
Vocabulary
- Artificial Intelligence
- Artificial intelligence is the field of making computer systems perform tasks that usually require human intelligence, such as recognizing patterns or making decisions.
- Machine Learning
- Machine learning is a method where a computer improves at a task by learning patterns from data instead of being programmed with every rule.
- Generative AI
- Generative AI is a kind of AI that produces new content, such as sentences, images, sounds, or code, based on patterns learned from examples.
- Training Data
- Training data is the collection of examples used to teach a machine learning model how patterns in inputs relate to outputs.
- Token
- A token is a small piece of text, such as a word, part of a word, number, or symbol, that a language model processes.
Common Mistakes to Avoid
- Assuming generative AI understands meaning exactly like a person is wrong because the model mainly predicts patterns from data rather than having human experiences or intentions.
- Trusting every AI answer without checking it is wrong because the model can produce false statements, outdated facts, or made-up sources that sound confident.
- Thinking the same prompt always gives the same response is wrong because many generative systems use probability, so small changes or random sampling can change the output.
- Ignoring the training data is wrong because biased, incomplete, or low-quality data can lead to biased, incomplete, or low-quality AI outputs.
Practice Questions
- 1 A text model chooses among three possible next tokens with probabilities 0.50, 0.30, and 0.20. What is the probability that it chooses either of the first two tokens?
- 2 A model is trained on 12,000 images. If 30 percent are pictures of cats, 45 percent are pictures of dogs, and the rest are pictures of birds, how many bird images are in the training set?
- 3 A student asks a generative AI tool to explain photosynthesis and receives a detailed answer. Describe two steps the student should take to decide whether the answer is reliable.