Researchers push back on AI doom narratives: real risks are human misuse, not rogue superintelligence

anshulkundaje · x · 2026-11-17

A debate on AI risk: giffmana argues the "superintelligence will self-replicate and enslave us" narrative is sci-fi fanfic — LLMs only generate tokens when model.forward() is called, and "quickly self-replicating" across clusters with terabytes of weights is impractical. The pressing risks are labor-market impact and cyber offense accessible to script kiddies. sirbayes agrees: AI deciding to kill everyone is BS; humans will use AI for bad (and good), just like any tech, only faster.

Related event: ML Researchers Push Back on AI Doom Narratives, Citing Human Misuse as Real Risk(3 posts)→

Original post →

More from AGI Musings

AGI Musings channel →