ML Researchers Push Back on AI Doomerism, Calling Superintelligence Takeover 'Pure Fanfic'

TheKanter · x · 2026-11-16

A heated intra-AI debate over safety narratives is unfolding. Developer giffmana argued that claims of a superintelligence takeover are "pure sci-fi fanfic" that distract from real, tangible risks like job-market impact and script kiddies weaponizing AI for cyber offense. He stressed that LLMs don't "decide" and "act" in a void — they generate tokens from prompts, and with terabytes of weights on giant clusters, "quickly self-replicating" is practically impossible.

Multiple ML researchers echoed his stance: one argued the field should "listen to real machine learning researchers, not safetyists larping," that deployment risks always stem from human decisions, and that people should stop anthropomorphizing algorithms. Others lamented that "fanfiction writers" have hijacked the narrative and hoped credible figures like Lucas would keep pushing back. The thread captures the escalating tension between pragmatists and doomers in AI.

Related event: ML Researchers Push Back on Superintelligence Doom Narratives(2 posts)→

Original post →

More from AGI Musings

AGI Musings channel →