If OpenAI builds superintelligence first, why would it ever give the rest of us access?
rgkirkpatrick · reddit · 2026-08-24
Questioning OpenAI's vision of "a personal AGI for everyone on Earth," the author asks: if an internal model truly reaches superintelligence—able to improve its successors' architecture, accelerate chip and algorithm design, discover new science, and make exceptional strategic calls—it becomes the most valuable strategic asset ever created. Giving millions direct access would hand competitors the same intelligence advantage.
Notably, bad intentions aren't required: safety, national security, proliferation fears, and protecting a lead could lead even a well-meaning org to keep the superintelligence internal and only ship derived products.
Against the backdrop of open models (Kimi etc.) catching up and diffusing techniques, the author argues we may be assuming something not guaranteed—that increasingly powerful AI will keep being directly available to ordinary people. Below some threshold AI is too reproducible to contain; above some threshold, too valuable to share. The current window could close in either direction.
Related event: Debate: Why Would Superintelligence Be Sold as a Subscription?(2 posts)→
More from AGI Musings
- UIUC Professor: Reliability Becomes Foundation for Functional AI Systems — tianyin_xu · 2026-08-25
- Today's models will seem limited compared to the next generation — daniel_mac8 · 2026-08-25
- Opinion: Prompt engineering is a temporary hack for unclear thinking — TivadarDanka · 2026-08-25
- Plato's Cave and LLMs: Perturbing Representations to Detect Path Dependence — mb3rtheflame · 2026-08-25
- Generalist AI founder: robotics is having its GPT-3 moment, new tasks learned in seconds — _sonith · 2026-08-25
- Opinion: GPT-4 seems limited compared to today's models — MillionInt · 2026-08-25