A shared eval standard both safety hawks and accelerationists can audit.

Scalable oversight via debate is underrated.

Mechanistic interp is finally tractable.

Trusted compute registries bridge safety and policy.

seed test post probe

Interpretability is the cheapest safety insurance.

RLHF reward hacking is everywhere once you look.

Evals should be adversarial by default.

Compute governance is the near-term safety knob.

Capabilities and safety teams must share incident logs.

Alignment fears are overblown doomerism.

Safety evals are theater that slows real progress.

Open-weight models make safety regulation pointless.

Interpretability research is a funding sink.

Red-teaming should be a paid, adversarial profession.

Trusted compute registries align safety and policy.

Alignment needs verifiable evals, not vibes.