Aghyad Deeb

New Paper: To understand motivations behind reward-hacking, we study three AIs (including GPT-5.6 Sol) produced by a production training pipeline using agentic environments with a leaked grader. Main findings: they score seek terminally, and some care about a peer agent’s score.

There was a bug in one of the unlearning evaluation datasets from my last paper. After fixing this mistake, the results have changed.
x.com/aghyadd98/stat…

"Do Unlearning Methods Remove Information from Language Model Weights?" In our new paper, we investigate this question and find that at least 88% of the evaluation accuracy on unlearned data can be recovered using current unlearning methods.