Castro argues braking AI also slows the research that makes it safer
castrotech · x · 2026-09-17
Castro posted a thread on the tension between AI safety and progress:
- The real challenge is building systems that become more capable and more reliable over time; hitting the brakes also slows the research, testing, and deployment that make AI safer.
- "I can imagine a catastrophic AI scenario" and "there's a 10% chance AI causes human extinction" are different claims — the first justifies serious safety research, the second requires evidence we simply don't have.
- The goal shouldn't be "safe AI instead of powerful AI" but safer, more powerful AI — investing in safety research, testing, and safeguards without braking the technology.
A clear anti-decelerationist viewpoint thread.
More from AGI Musings
- The part of AI nobody talks about: losing the joy of creating — BudgetCaterpillar946 · 2026-09-17
- DeepMind paper: Veo 3 shows emergent zero-shot reasoning, video models may become vision foundation models — RexDouglass · 2026-09-17
- Animals are plants' technology — so AI is humans': a viral framing — repligate · 2026-09-17
- Allie Miller's checklist for AI quitters: name real risks, not vague panic — alliekmiller · 2026-09-17
- dbasch dismisses AI doomer books as Michael Bay-style appeals to emotion, not calculation — dbasch · 2026-09-17
- Debate: LLMs are the art world's Copilot — intentionality and taste are the scarce inputs — akbirthko · 2026-09-17