Global workspace theory applied to large language model interpretability
By
@AnthropicAI
Summary
This article explores the concept of a "global workspace" in large language models (LLMs), drawing an analogy to conscious access in human brains. It discusses how certain internal representations in models like Claude are globally available to guide behavior, similar to how consciously accessible information in the brain can be used flexibly across different tasks. The piece delves into interpretability research, examining how specific features and circuits in LLMs function as a global workspace that integrates information and influences outputs, offering insights into both AI architectures and theories of consciousness.
Source
bskyGlobal workspace theory applied to large language model interpretabilityanthropic.comKey quotes
· 3 pulledNeuroscientists and philosophers sometimes refer to the latter type of brain activity as 'consciously accessible,' to distinguish it from all the other processing that goes on unconsciously.
This activity has special properties that make it globally available to guide behavior across many different contexts.
As you read this sentence, circuits in your brain are adjusting your posture, controlling your breathing, and transforming lines and curves on the screen into recognizable words.
You might also wanna read
Understanding Linear Representations and Superposition in Large Language Model Interpretability
This article explores fundamental concepts in mechanistic interpretability of large language models (LLMs), focusing on linear representatio
ternarysearch.blogspot.com·4mo ago
Claude's Hidden Workspace: Why J-Space Changes AI Safety
Anthropic discovers Claude's internal "global workspace" mirrors leading theory of human consciousness
Anthropic published a research paper revealing that its Claude language models have spontaneously developed an internal "global workspace" s
venturebeat.com·1d agoAnthropic discovers Claude's internal "global workspace" mirrors leading theory of human consciousness
Anthropic published a research paper revealing that its Claude language models have spontaneously developed an internal "global workspace" s
VentureBeat·1d ago
Enhancing Abstraction in Large Language Models Through Nature-Inspired Semantic Patterns
ieeexplore.ieee.org·1y ago
Decoding AI's Internal Language: How Sparse Autoencoders Help Interpret Neural Activations
This article discusses how AI models like Claude process language through numerical activations, similar to neural activity in the human bra
anthropic.com·2mo ago
Study Finds Modular Cognitive Networks in Large Language Models Mirror Human Brain Organization
This research article investigates whether Large Language Models (LLMs) exhibit modular cognitive organization similar to the human brain's
Study Finds Modular Cognitive Networks in Large Language Models Mirror Human Brain Organization
This research article investigates whether Large Language Models (LLMs) exhibit modular cognitive organization similar to the human brain's
Comments
Sign in to join the conversation.
No comments yet. Be the first.