Google

An LLM can be prompted with examples of problem decomposition to learn how to break down new, unseen problems. For instance, after being shown how other questions are broken down, the model can correctly generate the necessary sub-problems, such as 'When did the environmental study start?' and 'When did the environmental study end?', to solve a novel question about the study's duration.

Example of LLM Generating Sub-Problems for a Duration Question

Based on the provided scenario, analyze why the AI model failed to generate the correct sub-problems for the new question about the project's duration. What is the fundamental mismatch between the example provided and the new question posed?

Diagnosing a Flawed Problem Decomposition

An AI model is tasked with answering the question: 'For how many years was Queen Victoria the monarch of the United Kingdom?' To solve this, the model first needs to break the main question down into smaller, fact-based questions. Based on a strategy where problems are decomposed into their constituent parts, what are the two essential sub-questions the model must generate and answer first?

Generating Sub-Problems for a Duration Calculation

An AI model is provided with the following examples to learn how to break down a main question into a series of smaller, solvable sub-questions:

Example 1:
Main Question: What is the price difference between a laptop and a tablet?
Sub-questions: 
1. What is the price of the laptop?
2. What is the price of the tablet?

Example 2:
Main Question: How many more pages are in 'War and Peace' than in 'The Great Gatsby'?
Sub-questions:
1. How many pages are in 'War and Peace'?
2. How many pages are in 

Learn Before

Related