Sparring with Schmidhuber: alignment isn't a panacea, harden everything else instead

gandamu_ml · x · 2026-09-25

A follow-up in a spat involving Schmidhuber. The author rejects bullying regardless of political framing, then argues alignment is no panacea: even an aligned model doesn't stop training arbitrary models from getting ever easier. Better to make everything non-AI more robust; focusing only on the AI itself is "magical thinking."

Related event: Aftermath of Schmidhuber Debate: Alignment Is Not a Silver Bullet(2 posts)→

Original post →

More from AGI Musings

AGI Musings channel →