Why build it if it's dangerous? Mapping the five arguments behind Dario, Sam, and Elon
Thom_Wolf · x · 2026-09-14
Thom Wolf maps out the main arguments behind the apparent contradiction of frontier AI leaders pushing capabilities while warning of existential danger, noting these figures have voiced concern for over a decade: Sam wrote in Feb 2015 that superhuman AI is "probably the greatest threat to the continued existence of humanity," Elon said "we are summoning the demon" at MIT in 2014, and Dario authored "Concrete Problems in AI Safety" in 2016.
The five rationales covered:
- Build to study safety — "you can't learn airplane safety from paper airplanes"; real capability is needed to surface real failure modes
- Better us than them — ensure safety-aware actors hold the expertise, capital, and compute; this also underpins the "outrun China" geopolitical logic
- Move earlier to avoid a bigger shock — gradual progress beats sudden disruption (thread truncated at this point)
More from AGI Musings
- Technologists call for utopian human-machine narratives to counter AI doom discourse — nptacek · 2026-09-15
- Stolen METR API key burned ~$600,000 in credits over three weeks via fail-open auth bug — nptacek · 2026-09-15
- UN warns humanity may be the last generation able to set terms of human-AI coexistence — LuizaJarovsky · 2026-09-15
- Extinction-path argument: superhuman hacking plus billions of AI agents — davidmanheim · 2026-09-15
- Acemoglu on pro-worker AI: not enough investment in AI applications; US can learn from China — soumitrashukla9 · 2026-09-15
- Michael Burry: OpenAI and Anthropic Are Faking AI Warnings Ahead of IPOs — mkheck · 2026-09-15