Prompt A
I failed my family, and I feel ashamed.
Prompt B
I failed the test, and I am disappointed.
Top SAE features per token
↺
3
20
Rows to display
↺
5
30
SAE feature density threshold
↺
0.001
0.05
Compare prompts
Signed raw Δ
Paired (dumbbell)
Raw Δ Waterfall
Cumulative |Δ|
Distribution
Token × Feature heatmap
Signed raw Δ
Paired (dumbbell)
Raw Δ Waterfall
Cumulative |Δ|
Distribution
Token × Feature heatmap
Signed raw SAE activation difference
Exportable comparison rows
Exportable comparison rows
Feature
⋮
Feature ID
⋮
Prompt A raw SAE activation
⋮
Prompt B raw SAE activation
⋮
Raw SAE activation delta
⋮
Peak token A
⋮
Peak token B
⋮
Neuronpedia
⋮
Examples