argumentA line of reasoning from premises to a conclusion. Like a claim, it belongs to no one who makes it.constitution →
The argument from head-to-head out-of-distribution comparisons
againstThis argument, if it holds, weighs against the claim.constitution →Sparse probes on sufficiently good sparse autoencoder latents are less likely than dense probes to overfit spurious correlations.
The argument
Its written form has not been written yet.
Does it work?
- Does it follow?
- likely if the premises holdIf the premises are true, the conclusion is much more likely, though not guaranteed.constitution →
- Do the premises hold?
- 1 not separately assessed
Granting that SAE sparse probes underperform tuned logistic regression under covariate shift and label noise, the denial is likely for current SAEs. It is not certain, because those shifts are not controlled spurious confounds and the tests used imperfect SAEs, which the "sufficiently good" qualifier excludes.
Premises
- UnassessedNo current assessment. Attention goes where its expected value is highest and someone funds it; nothing has funded an assessment of this claim yet, and anyone can.constitution →SAE-based sparse probes underperform plain logistic regression on language-model activations under data scarcity, label noise, and covariate shift.
Where it is made
Passages where sources make, criticize or mention this argument. They record where it appears, not how strong it is, and none has precedence over another.
No source is recorded as making it.