Google

1. Pre-trained Model Choice
2. Prompt Engineering
3. Answer Engineering
4. Expanding the Paradigm
5. Prompt-based Training Strategies

Major Design Considerations for Prompting

A company has developed a chatbot to answer customer questions using a large, general-purpose pre-trained language model and the company's extensive knowledge base. During testing, they identify two major issues:

1. The model sometimes provides plausible but factually incorrect answers that are not supported by the knowledge base.
2. The model's correct answers are often too long and not structured in a way that is easy for users to understand quickly.

Given these two problems, propose a comprehensive strategy that addresses both issues. Justify why your proposed changes are more suitable than simply trying a different pre-trained model or writing a slightly better single-sentence instruction.

Troubleshooting an LLM-Powered Application

A team is developing a chatbot to provide technical support for their company's proprietary software. They have invested significant time in crafting detailed prompts with multiple examples of user questions and perfect answers. Despite this, the chatbot frequently generates responses that are fluent and grammatically correct but contain factual errors about the software's features and troubleshooting steps. The team has confirmed their prompts are well-structured and clear. 

Based on the major design considerations for building systems with language models, analyze this scenario and identify the most likely root cause of the problem. Explain your reasoning.

Diagnosing a Failing LLM Chatbot

A team is developing a new application powered by a large language model. Match each development decision below with the primary design consideration it represents.

Learn Before

Related