iamtrask: alignment is a data problem, and humanity is '100% gonna survive and thrive'
iamtrask · x · 2026-09-10
OpenMined founder iamtrask says he's incredibly optimistic about AI safety progress in 2026: the alarm is working and humanity will survive. His core takes: alignment is a data problem (data cleaning, RLHF tools exist, just not deployed right); oversight and governance is a cryptography-plus-social-institutions problem with all ingredients in place; extreme positions serve a balancing role but we'll settle in the middle; the biggest bottleneck is communication bandwidth between communities that don't trust each other — and it's improving.
More from AGI Musings
- New AI whistleblower draws Blake Lemoine comparisons: the engineer fired for claiming AI was sentient — IsForAt · 2026-09-10
- MacAskill & Moorhouse: intelligence explosion could compress a century of progress into a decade — zetalyrae · 2026-09-10
- OpenAI's new model, in training since Aug 28, reportedly beat GPT-6-Astra in just one week — alexcovo_eth · 2026-09-10
- Beginner's paradox in the AI era: learn fundamentals or just ship with AI — HankYeomans · 2026-09-10
- AI Safety Researcher David Krueger: Gambling With Your Life Is a Sick Culture — DavidSKrueger · 2026-09-10
- CRM Won't Be Ripped Out, But Its Agent-Layer Risk Is Real, Investors Argue — matt_slotnick · 2026-09-10