Ben Lorica Previews the Frontier AI Regulation Debate: You Can't Ban RSI Without Defining It
bigdata · x · 2026-11-02
Ben Lorica previews the frontier AI lab regulation debate ahead of the US midterms, with a skeptical lens (he self-identifies as low P(doom) and favors governing AI via existing sector-specific laws).
Key points:
- Proposals to pause recursive self-improvement or ban superintelligence lack enforceable definitions: he distinguishes an AI system making bounded improvements from one improving the process that generates further improvements. Does a coding agent improving a training pipeline count? An agent running experiments and picking winners? Without a clear definition, a ban's scope is unknowable.
- Requiring labs to evaluate internal models before using them to develop successors is more practical in principle, but who sets the criteria and who certifies a model is safe for that work? Lab self-assessment likely isn't enough.
- Mandating 3-9 months of public deployment before internal research use is unconvincing: public use may not reveal the risks of handing a model to internal researchers.
Context: if Democrats win at least one chamber or several governorships, frontier lab regulation proposals could get a more serious hearing.
More from AGI Musings
- Anthropic S-1 filing: may imminently cross its RSP threshold for AI-driven accelerated progress — kimmonismus · 2026-11-03
- Sam Altman: AI is now mostly smarter than our smartest scientists — rohanpaul_ai · 2026-11-03
- Frank Calegari essay: mathematics has changed with AI — mathematicians should act like it — littmath · 2026-11-03
- Robotics is having its GPT-2 moment, says State of AI report author Nathan Benaich — nathanbenaich · 2026-11-03
- Daniel Faggella makes the cosmist case against anthropocentrism: the flame, not the torch — danfaggella · 2026-11-02
- Sabbatical researcher maps 5 theses: bifurcated software, personal agents, attention repricing — manosaie · 2026-11-02