Jack Clark (Import AI)

Jack Clark's Import AI covers research on AI oversight difficulty, new scaling law findings for protein folding models, and economic frameworks for valuing extinction-level AI risk.

Jack Clark's Import AI #458 centers on a lengthy essay reckoning with AI's long-term trajectory alongside a speculative singularity narrative.

Jack Clark's Import AI #457 covers a newly discovered AI-based cyberweapon (AI Stuxnet), an unusual Muon optimizer result, and recent work on positive AI alignment.

Jack Clark's Import AI #456 covers recursive self-improvement and economic growth, a novel "radical optionality" regulatory approach, and a new neural computer architecture.

Jack Clark's Import AI #455 argues AI systems are approaching the threshold where they can meaningfully contribute to their own training and research pipelines.

Jack Clark's Import AI #454 covers efforts to automate alignment research, a safety study of a Chinese model, and Huawei's HiFloat4 training data format.

Jack Clark's Import AI covers new research on breaking AI agents, the MirrorCode project, and ten perspectives on AI's gradual disempowerment of human institutions.

Jack Clark's Import AI covers scaling laws for AI-enabled cyberwarfare, the rising automation of economic tasks, and puzzling patterns in AI-based GDP forecasting.