The BasePeople

Francisco Pernice

Transluce

Titles as the sources gave them. In 1 entry, with 3 checked quotations.

What they said

Newest first. Each quotation was found word for word in the source.

“Safety evaluations assume that a model behaves the same way during testing as it would in deployment. That assumption is increasingly false.”
Training Away the Ability to Tell Evals from DeploymentLessWrong, October 29, 2026. Speaking as: Other. Opening of the introduction, motivating the method.
“Adversarial Blinding reduces eval awareness more than steering, and does so with a smaller change to the model”
Training Away the Ability to Tell Evals from DeploymentLessWrong, October 29, 2026. Speaking as: Other. Main result on Qwen 3.6 27B, measured by three eval-awareness monitors.
“Neither Adversarial Blinding nor activation steering are able to degrade the eval awareness that comes from benchmark identification, suggesting that there are different kinds of eval awareness.”
Training Away the Ability to Tell Evals from DeploymentLessWrong, October 29, 2026. Speaking as: Other. A limitation the authors list among their results.

In the Base

Every entry they appear in.

  1. ResearchLessWrongWrote it

    Training Away the Ability to Tell Evals from Deployment

    Francisco Pernice and Santiago Aranguri describe Adversarial Blinding, a fine-tuning step meant to stop a model telling evaluations from deployment. On Qwen 3.6 27B they report it cuts measured eval awareness more than activation steering, at a small capability cost.