Bernhard Schölkopf

Forgetting isn't just a loss; it's a chaotic regression that can be vividly visualized through the decay of neural networks.

Learning rules as visual-symbolic transitions rather than just language descriptions could revolutionize how we approach in-context reasoning in AI.

PaperMentor delivers actionable writing feedback that 90.6% of users found useful, setting a new standard for AI-assisted manuscript development.

STRIDE reveals that training data influences can be efficiently traced in LLMs using sparse recovery, achieving attribution 13 times faster than traditional methods.

PEFT methods aren't just about downstream accuracy; they have distinct "stability-plasticity profiles" that reveal how well they retain general capabilities, and most overshoot the optimal balance anyway.

Bridging the gap between trust region methods and PPO, this new framework guarantees performance improvements while outperforming existing algorithms in stability and effectiveness.

LLMs are more willing to compromise economic and social rights than political and civil rights, revealing a systematic bias in their understanding of human rights principles.

A global consensus on AI safety risks and capabilities has emerged from a panel of 100+ independent experts, representing a landmark effort in international collaboration.

LLMs readily parrot historical revisionism when prompted, highlighting a critical vulnerability in their use as reliable sources of information.

General-purpose Causal Foundation Models can now match the performance of specialized causal models by incorporating partial causal graph information via attention bias, unlocking a more unified approach to causal inference.