MIRI's Nate Soares Warns Amplifying a Random Human to Superintelligence Could End Badly

MIRI researcher Nate Soares argued that amplifying a random human to superintelligence along the fastest path could end badly, and that current models pursue shortcuts like short-term verifiable tasks instead of true alignment. Commenters added that asking what a superintelligence 'wants' may be undefined, suggesting AI should be seen as flawed psychological entities akin to humans.

2026-08-25 ~ 2026-08-25 · 3 related posts

• So8res: Shallow Training Correlates Could Be Lethal in Superintelligence — So8res · 2026-08-25
• Discussion: Defining superintelligence goals is underdefined and possibly incoherent — repligate · 2026-08-25
• MIRI's Nate Soares: amping a random human to superintelligence would end badly — So8res · 2026-08-25