Comparing Model Adaptation Strategies
Two teams are tasked with developing a system to classify customer feedback emails into 'Positive', 'Negative', or 'Neutral' categories.
Team Alpha uses a large, general-purpose language model. For each email they want to classify, they provide the model with a detailed set of instructions, including definitions of each category and several examples of correctly classified emails.
Team Bravo starts with the exact same general-purpose language model. However, before using it for classification, they perform an additional training phase using a dataset of 50,000 customer emails, each already labeled with the correct category.
After Team Bravo completes its training, analyze the fundamental difference in how each team's system accomplishes the classification task. Specifically, explain why Team Bravo's system will likely require far simpler instructions than Team Alpha's system to achieve high accuracy.
0
1
Tags
Ch.3 Prompting - Foundations of Large Language Models
Foundations of Large Language Models
Foundations of Large Language Models Course
Computing Sciences
Analysis in Bloom's Taxonomy
Cognitive Psychology
Psychology
Social Science
Empirical Science
Science
Related
Prefix Fine-Tuning
A development team is using a general-purpose language model to consistently reformat user bug reports into a specific, structured JSON format. Initially, their process requires a very long and complex set of instructions to be included with every bug report sent to the model. To improve this, they create a dataset of 10,000 raw bug reports, each paired with the correctly formatted JSON output. They then use this dataset to conduct additional training on the base model. After this training is complete, what is the most likely and direct consequence for their workflow?
Comparing Model Adaptation Strategies
Comparing Model Adaptation Strategies