Open-weight LLMs memorize books far more than previously believed. Memorization varies by model family, model size, and book. In extreme cases, entire books are memorized, and we can generate them eff...

Why is Gaussian DP a much better compact representation of privacy guarantees than (ε,δ)-DP.

First I will consider a concrete problem for which I sought to use correlation; then, being not much of a statistician, I will refresh my memory of how it works; finally, I will consider the history o...

Current practices for reporting the level of differential privacy (DP) protection for machine learning (ML) algorithms such as DP-SGD provide an incomplete and potentially misleading picture of the pr...

Numeric computation of the privacy parameter in Gaussian Differential Privacy - interpretable-dp/gdpnum

In differentially private (DP) machine learning, the privacy guarantees of DP mechanisms are often reported and compared on the basis of a single $(\varepsilon, δ)$-pair. This practice overlooks that ...

Differential privacy (DP) is a widely used approach for mitigating privacy risks when training machine learning models on sensitive data. DP mechanisms add noise during training to limit the risk of i...

Experts in AI call on the President of the European Commission to retract unscientific AI hype statement she made in the budget speech.

Since the widespread release of ChatGPT in December of 2022, AI has taken over much of the world by storm – including academia. Most of this happened with very little pushback, despite a myriad of iss...

We know copyright

Paul Meehl's foundational work "Clinical versus Statistical Prediction," provided early theoretical justification and empirical evidence of the superiority of statistical methods over clinical judgmen...

by John Perry Barlow Governments of the Industrial World, you weary giants of flesh and steel, I come from Cyberspace, the new home of Mind. On behalf of the future, I ask you of the past to leave us ...