LIBERO-PeRM: Benchmarking Personalized Robotic Manipulation

New research challenges the superposition theory of polysemanticity, suggesting neural networks are more interpretable than previously believed through recursive generalization patterns.

Nick Cammarata argues neural networks are already clean and interpretable, challenging previous pessimism about mechanistic interpretability.

A new paper shows that fine-tuning AI models to give bad advice only creates narrow, fragile 'bad guy' personas that break easily with minor prompt changes.

Chinese AI safety research faces significant challenges with limited frontier risk awareness and systemic headwinds according to field research.

Researchers documented mercenary hackers using targeted ads to deliver zero-click spyware that infected Al Jazeera journalists' devices without any user interaction.