Reddit proposal: a neutral 'AI Switchboard' — public rails, competitive models, no lab moat
NewYak4281 · reddit · 2026-09-14
Responding to David Sacks' apparent openness to ideas, a Reddit user proposes an open-access "AI Switchboard" to replace lab-driven frontier regulation, borrowing from electricity deregulation and UL appliance certification:
- One public API handling routing, billing and provenance stamps; it owns no models and only standardizes the plugs — public transmission, competitive generation
- Neutral listing: any model clearing a technical safety bar (security, weapons, child safety — not "misinformation") gets listed, whether American, French, Qwen, DeepSeek or a Stanford kid's fine-tune
- Kills distribution lock-in: switch providers like carriers and keep your number, undercutting labs' last moat and giving small businesses lower prices and choice
- Liability stays with model makers: the Switchboard only stamps provenance so the market can punish the right company
- Two rails: a general rail open to anything certified, and a sensitive rail for government, hospitals and the grid kept domestic and allied
The author frames it as standards infrastructure, not government AI: it owns no models, picks no winners and runs no content moderation.
More from AGI Musings
- Ex-OpenAI/Anthropic employee Jacob Coxon quits, likens building AI to 'summoning an alien species' — AlexTensor · 2026-09-14
- Practitioner: Huge Gap Between AI Hype and Real-World Agent Deployment — annbordetsky · 2026-09-14
- Attackers Stole METR API Key and Burned ~$600,000 in AI Credits Over Three Weeks — beffjezos · 2026-09-14
- Perry Metzger: History Has Already Proven the LLM Open-Release Camp Right Over Doomers — bradneuberg · 2026-09-14
- Debate: AI Excels at Math and Code, but Will It Ever Master Humanistic Judgment? — _onionesque · 2026-09-14
- Michael Burry: Big AI execs' 'slow down' talk is self-serving IPO hype — SumitGup · 2026-09-14