Claims Browser

Existing post-hoc explainability methods for time series forecasting often ignore temporal dependence and fail to provide horizon-specific explanations

"existing post-hoc methods often ignore temporal dependence and fail to provide horizon-specific explanations"

Global workspace theory lacks a formal criterion for identifying the mechanism that enables conscious access

"Global workspace theory explains conscious access as the broadcasting of selected information to the rest of the network, but it lacks a formal criterion for identifying the mechanism that enables this access."

J-lens readouts in early layers are often noisy and largely uninterpretable

"we find readouts in early layers to often be noisy and largely uninterpretable"

There is a lack of controllable and attributable methods for analyzing how language models resolve conflicts between competing specifications

Existing gradient-based explainability methods struggle to provide global insights into what specifically drives similarity in regions of an embedding space

Existing representation engineering methods are evaluated on paper-specific synthetic data that is difficult to compare or reproduce and may reflect surface patterns rather than capabilities

Existing sufficiency-oriented explanation methods can assign high importance to spurious subsequences that support predictions without being essential to the model's decision.

Previous GNN explanation methods neglect synergistic effects among edges, which are crucial for accurately characterizing edge importance

Existing work on LLM self-explanation faithfulness focuses on evaluation or inference-time prompting but does not provide a mechanism to directly optimize model parameters for faithful self-explanations

The Transformer became dominant due to hardware constraints rather than principled architectural choices

The uniform Transformer architecture is a structural error compared to the brain's mosaic design with functionally-specialized regions

Neural network-based operator learning architectures are opaque models that obscure the reasoning behind their predictions

Existing multimodal knowledge editors rarely control the semantic boundary of each edit, leading to a scope gap where instance-level success doesn't guarantee proper generalization

Standard language models fail to provide traceable training data influence because their dense network pathways distribute influence across parameters