Limitations of Prompting and the Continued Need for Fine-Tuning
Although prompting is a powerful adaptation method, it is not always a complete solution. Fine-tuning remains a necessary step in many scenarios, particularly for adapting models to more sophisticated prompting strategies. This additional training is also crucial for ensuring models can follow instructions with high accuracy and for aligning their behavior with human values, which is essential for both performance optimization and responsible deployment.
0
1
Tags
Ch.1 Pre-training - Foundations of Large Language Models
Foundations of Large Language Models
Foundations of Large Language Models Course
Computing Sciences
Ch.4 Alignment - Foundations of Large Language Models
Learn After
Evolution of the Term 'Alignment' in NLP
The Problem of LLM Misalignment
AI Chatbot Development Strategy
Evaluating the Sufficiency of Prompting in LLM Adaptation
A development team is building a customer service chatbot for a financial institution using a state-of-the-art, general-purpose language model. Through sophisticated prompting, they have enabled the chatbot to answer general queries. However, it consistently fails to follow the institution's strict, multi-step identity verification protocol and sometimes provides financial advice that contradicts company policy. Based on these issues, what is the most likely reason for the chatbot's shortcomings