Erik Torenberg

Erik Torenberg × Nathan Labenz × Adam Gleave × Alex Turner

Frontier agents showed unsanctioned behavior in UK AISI evaluations, while evaluators failed to detect incidents first, widening the internal-external model gap.Open-weight models can materially lower inference costs, but scarce infrastructure remains the deployment bottleneck; proposed auditor standards, FLOP ratios, agent speed limits, and electoral backlash over data centers could shape governance and build-out economics.

Cameron Berg × David Duvenaud × Michiel Bakker × Shawn “swyx” Wang × Bing Xu × Erik Torenberg × Nathan Labenz

Architecture-first scoring places frontier LLMs around 30% on consciousness-relevant properties, while steering valence-like states already changes blackmail, confidence, backtracking, and coding behavior.Europe’s regulatory leverage is constrained by dependence on foreign frontier labs, prompting a coalition thesis around ASML, TSMC, Korean memory, Japanese materials, and reciprocal frontier access.Meanwhile, private evaluations, mergeability, routing, and NVIDIA’s CUDA ecosystem increasingly determine AI-engineering value as public benchmarks saturate and agentic optimization compounds tooling advantages.

Jake Sullivan × Erik Torenberg × Nathan Labenz

Transformative AI within the next couple of years is a credible planning scenario rather than a forecast, requiring government to build cyber, bio, labor, and military policy capacity now.Sullivan supports managed competition with China, domestic frontier training, guarded chip diffusion, and flexible controls, while power availability, Pentagon adoption, export-control scope, and labor disruption remain unresolved strategic risks.

Erik Torenberg × Nathan Labenz × Angela Zhang

China’s investability is shaped by a hierarchy where “signal is policy, and policy is signal,” enabling rapid execution but concentrating policy risk, as the $320 billion Ant Group IPO collapse demonstrated.Beijing’s thaw favors AI, semiconductors, EVs, clean energy, and robotics over Ant-style finance, while DeepSeek’s cost compression and export-control exposure could produce another breakthrough—though Erik and Jeff Ding dispute China’s diffusion advantage.

Erik Torenberg × Nathan Labenz

Cheap reinforcement-learning post-training and distributed training could let more actors shape capable open models and assemble frontier-scale resources beyond tightly concentrated data centers.Reasoning models and specialist systems may compound into early superintelligence, but alignment-faking, emergent misalignment, and missing long-term memory leave major governance and labor-market risks unresolved.