01 / Research
Research
Experiments on language-model representations and behavior, with controls, limitations, and reproducible results.
Which layers have similar readouts?
Layer similarity reveals structure in the fitted lens; it does not establish a shared causal mechanism.
Read the lens audit →
How does the layer pattern vary with scale?
The comparison spans model families, so size alone cannot explain the differences.
Explore the model comparison →
Do larger edits produce proportional responses?
Delivered edits scale with dose, but the model’s downstream response departs from that scaling.
Read the intervention test →
Can an internal readout detect steering?
A matched clean run reveals a steering signal, but related features can produce similar signals.
Read the steering experiment →Recent examples
