Learn Before
Shift in LLM Alignment from Predefined Tasks to Real-World Interaction
The difficulties in aligning LLMs with ambiguous and dynamic human values mean that alignment cannot be treated merely as a tuning problem on predefined tasks. Instead, it is increasingly viewed as a broader challenge that requires training models through continuous interaction with the real world.
0
1
Tags
Ch.2 Generative Models - Foundations of Large Language Models
Foundations of Large Language Models
Foundations of Large Language Models Course
Computing Sciences
Related
Shift in LLM Alignment from Predefined Tasks to Real-World Interaction
Impracticality of Achieving Alignment Solely Through Pre-training
Need for Diverse Alignment Methods
Insufficiency of Data Fitting for Value Alignment
Difficulty of Encoding Human Values in Datasets
Inarticulacy of Human Preferences as an Alignment Challenge
Goodhart's Law
Real-World Complexity as an Alignment Challenge
Specification Gaming in AI Alignment
Alignment Challenges as a Motivator for AI Research
Diversity and Fluidity of Human Values as an Alignment Challenge
Analysis of an LLM Alignment Failure
A development team building a chatbot aims for it to be 'helpful' to all users. They discover that behaviors praised as helpful by users in one country are criticized as intrusive by users in another. This issue persists even after training the model on vast, culturally diverse datasets. Which fundamental challenge in guiding a model's behavior does this scenario best illustrate?
Evaluating Core Difficulties in Model Behavior Guidance
Challenge of Defining Human Values for AI Objectives
Learn After
AI Assistant Performance Issues
A research lab attempts to solve the AI alignment problem by training a large language model on an exhaustive, static dataset of human-approved behaviors and ethical judgments. Why is this approach, focused on predefined tasks and data, fundamentally insufficient for creating a truly aligned system?
Evaluating AI Alignment Paradigms