Language models are sequential processors, but long-horizon agency requires external information and computation beyond model weights and active context. Prime Agent is an open-source harness for…

Novel deep learning architectures are increasingly being applied to biological data, including genetic sequences. These models, referred to as genomic language models (gLMs), have demonstrated…

"AI Control" is the research agenda surrounding the set of techniques we can use to mitigate the harmful capabilities of AI to get useful work, even if it is…

Freed from intelligibility and aesthetics, AI designs faster

Right now, across the world, AI agents are repeating the same absurd act: to read one document, they each recompute it from scratch. Every agent re-runs prefill, the most compute-intensive step a…

Reinforcement learning agents can exploit misspecified reward signals to achieve high apparent returns while failing on the intended objective, a failure mode known as reward hacking. Existing…

What David Graeber’s Debt: The First 5,000 Years reveals about gratitude, interdependence, and why UBI is the smallest acknowledgment civilization can make to the unpayable debt of being alive

Evan et al argue for developing "model organisms of misalignment" - AI systems deliberately designed to exhibit concerning behaviors like deception o…