Paper Feed: May 2026

Training and Interpretability

• Synthetic Data for any Differentiable Target
  Tristan Thrush, Sung Min Park, Herman Brunborg, [...], Tatsunori Hashimoto (2026)
• Mechanisms of Introspective Awareness
  Uzay Macar, Li Yang, Atticus Wang, Peter Wallich, Emmanuel Ameisen, Jack Lindsey (2026)
• Subliminal Effects in Your Data: A General Mechanism via Log-Linearity
  Ishaq Aden-Ali, Noah Golowich, Allen Liu, Abhishek Shetty, Ankur Moitra, Nika Haghtalab (2026)
• Disentangling MLP Neuron Weights in Vocabulary Space
  Asaf Avrahamy, Yoav Gur-Arieh, Mor Geva (2026)

Misalignment and Generalization

• Comment on: Current AIs seem pretty misaligned to me
  Sam Marks (2026)
  
  Some notes
  See also the Greenblatt follow-up.
• From personas to intentions: towards a science of motivations for AI models
  David Africa, Jacob Pfau (2026)
• How do LLMs generalize when we do training that is intuitively compatible with two off-distribution behaviors?
  Dylan Xu, Alek Westover, Vivek Hebbar, [...], Julian Stastny (2026)
• Blind Refusal: Language Models Refuse to Help Users Evade Unjust, Absurd, and Illegitimate Rules
  Cameron Pattison, Lorenzo Manuali, Seth Lazar (2026)

Security, Governance, and Agents

• Coding Agents Are Changing the Biosecurity Risk Landscape
  Luca Righetti, Kamile Lukosiute, James Black (2026)
• LinuxArena: A Control Setting for AI Agents in Live Production Software Environments
  Tyler Tracy, Ram Potham, Nick Kuhn, [...], Aryan Bhatt (2026)

AI Economics and Forecasting

• WAGE-Bench: Measuring the Economic Value of AI in Real Work
  Kris Gulati (2026)
• The least understood driver of AI progress
  Anson Ho (2026)