- (a) A generative model writes a brand new short story after studying many of them. A discriminative model simply looks at a new text and decides if it’s a story or a poem.
- (b) A generative model recognizes if a voice belongs to a male or female. A discriminative model creates a unique, human-like voice that’s never existed.
- (c) A generative model supports you in editing a picture. A discriminative model creates a picture based on the prompt given to it.
- (d) A generative model composes a fresh new tune. A discriminative model corrects the lyrics of a song you give it as input prompt.
Correct Answer: (a)
A generative model learns how data is formed and is capable of producing new data based on its learnings. While, a discriminative model classifies or separates data into groups and does not create any new data.
As you might have noticed, all other options except option (a), do not comply with the definition of both models.
Only in option (a), the generative model is being used for generating something (a story in this case), and the discriminative model is being used for classification of texts into a poem or story.
Let’s understand the AI models discussed in the question in more detail:
Generative AI Models
This model focuses on understanding the entire spectrum of the data provided to it and then generating new data based on its learnings. It analyzes the input data and based on the prompts given by the user, it is capable of generating new data that is based on the data provided at the time of input.
If this sounds confusing, an example of a Generative AI model will surely help.
GPT is one of the most famous examples of generative AI models. It is trained on large amounts of text-based data.
GPT does not directly store this data into its memory. Rather, it analyzes the input data, recognizes patterns in it, understands its structure, and learns from all of this.
Later, when a user passes a prompt, GPT acts on it based on its learnings and generates data that feels authentic and unique.
Discriminative AI Models
These types of models have more to do with learning the boundary between different classes of data and classifying the input into the segments where it fits the best. Discriminative models are majorly used for categorization of data and are not concerned with the generation part.
Logistic Regression can be given as an example for this model. It focuses on answering questions like: ‘Given this piece of text, determine in which language it is written?’.
Rather than focusing on how the text was written, it’ll analyze patterns, special characters, and structure of the text. Based on these, it’ll classify the text further into a language category.
In conclusion, we can say that Generative AI is like a novelist, while Discriminative AI is like a critic reader who decides in which category, fictional, inspirational, thriller, crime, etc., the novel will go.