Someone is wrong on the internet (AGI Doom edition)

Article reception and framing

• Several commenters criticize the blog post’s tone (ad hominems, contempt) and claim it misrepresents AI risk concerns as “LLMs becoming conscious and killing everyone.”
• Others enjoy it as a cathartic rant and appreciate its mockery of certain online rationalist communities, even while disagreeing with the substance.
• Some argue the piece ignores that text alone already encodes a lot of real-world structure, as evidenced by LLM capabilities.

Nature and plausibility of AGI risk

• One camp sees “AGI doom” as overblown, likening it to a quasi-religious or Maxwell’s Demon–style thought experiment, and stresses fundamental physical and information-theoretic limits.
• Others argue that:
  • General or superhuman AI is clearly possible “in theory” (brains are physics, so simulatable).
  • Once you assume an agent far beyond human capability, with goals not perfectly aligned to human survival, catastrophic conflict over resources is plausible.
  • Instrumental convergence (e.g., power accumulation) and principal–agent problems could make dangerous outcomes common even without explicit malice.

Physical vs digital threat models

• Skeptics emphasize:
  • No direct internet connection to nuclear launch systems and strong human command chains.
  • Serious practical barriers to embodied, human-level robotics (energy storage, real-world dexterity).
• Others counter:
  • AI can cause major harm purely online (financial systems, infrastructure, surveillance, internet “war”).
  • Autonomous weapons, “human-free militaries,” and AI-assisted bioweapons (DNA printers, lab leaks, “test in production”) are more realistic than sci‑fi robot uprisings.

Human agency, incentives, and governance

• Broad agreement that humans deploying and weaponizing AI are central to the risk.
• Concerns include:
  • Mis-specified objectives and bugs in powerful systems.
  • Corporations and states using AI for profit, repression, or war.
  • Regulatory capture by current AI leaders; using distant x‑risk narratives to justify locking down the field.

Near-term socioeconomic and political risks

• Many argue job displacement and productivity gains accruing only to capital are already happening and are more urgent than extinction scenarios.
• Fears of an equilibrium where a small elite controls automated production and security, leaving most people economically “not worth feeding,” even without explicit extermination.
• Others note historical patterns: inequality and declining living standards can drive nationalism, conflict, and authoritarian responses.

Manipulation, addiction, and information harms

• Strong concern that AI-boosted recommender systems and generative content will act as extremely powerful “skinner boxes,” hijacking human attention and agency.
• Some see this as the primary present-day danger: governments and corporations using AI to shape behavior, filter reality, and enshittify digital environments.

Technical pathways to AGI

• Several commenters argue LLMs plus feedback loops, sensors, and goals could be enough for practical AGI; hallucinations are compared to normal human perception errors corrected by feedback.
• Others emphasize that real-world experimentation and embodiment remain hard, and current RL/game-based approaches have largely been eclipsed by text-trained models.

Role of science fiction and public perception

• Commenters note that public and even expert intuitions are heavily shaped by sci‑fi tropes (Skynet, “Her,” etc.), often poorly matched to actual systems.
• Some warn that dismissive takes like the article’s may themselves be evidence of dangerous overconfidence and normalcy bias.