GWT-2: Limited capacity workspace, entailing a bottleneck and a selective attention mechanism

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• GWT-1: Multiple specialised systems capable of operating in parallel (modules)concept0.820
  Indicator from GWT: existence of specialized parallel modules.
• GWT-4: State-dependent attention enabling sequential querying of modules for complex tasksconcept0.805
  Indicator: state-dependent attention and sequential module recruitment for System 2 thought.
• GWT-3: Global broadcast: availability of information in the workspace to all modulesconcept0.792
  Indicator: global broadcast to all modules.
• A system with high integrated information but no goals, or a global workspace broadcasting non-evaluative content, would not be conscious on the present accountclaim0.735
  Point of genuine disagreement with IIT and GWT
• Perceiver architecture is some way from being a full implementation of GWT.claim0.734
  Case study conclusion.
• Treating GPT as an unsupervised implementation of a supervised learner leads to systematic underestimation of capabilities.claim0.721
  Critique of the oracle/supervised frame.
• Interpretability in the Wild: A Circuit for Indirect Object Identification in GPT-2 Small (Wang et al., 2023)concept0.713
  Cited as causal intervention methodology precedent for this paper's ablation approach
• Attention is a generalization of convolution; all convolutions can be expressed as tensor products of fixed relative position attention patterns and weight matricesclaim0.713
  Mathematical equivalence showing the relationship between attention mechanisms and convolutional operations