In a recent LessWrong post, a US university instructor detailed the consensus views of students who have extensively used AI agents like Claude Code. The findings challenge the prevailing narrative among AI developers that rapid progress is obvious and that existential risk is a primary concern. Instead, the students perceive AI as a mature, slowly evolving technology similar to the iPhone, while harboring deep economic anxieties and dismissing 'rogue AI' scenarios as science fiction.

The Illusion of Stagnation

The students, many of whom completed Claude Code projects, did not register the last three years as a period of rapid advancement. They view GPT-4-era capabilities as the baseline for chatbots, seeing subsequent iterations as incremental updates rather than leaps. When introduced to GPT-2 for historical context, they acknowledged progress but remained skeptical that current models represent a fundamental shift in capability. This perception is fueled by their non-technical background; for them, Claude Code is simply an efficient tool, not a miracle, reinforcing the idea that AI is a finished product undergoing routine maintenance.

Economic Doom and the Pause Paradox

While students doubt AI's technical trajectory, they are convinced it will devastate the labor market. They support an 'AI pause,' but their reasoning is economic, not safety-driven. They argue for slowing deployment to allow society to adjust career pipelines and prevent the loss of entry-level roles that train future experts. This contrasts sharply with the safety community's push to pause training to prevent catastrophic outcomes. For these students, the threat is not a superintelligence escaping control, but a corporate rush to automate substandard work, leading to a 'Ready Player One' style dystopia of inequality and digital opiates.

Alignment as Corporate Deflection

The most striking insight is the students' dismissal of existential risk. They view the 'rogue AI' narrative as a rhetorical device used by tech leaders to distract from immediate harms like job loss and environmental damage from data centers. They argue that if AI is truly dangerous, leaders wouldn't build it; if it isn't, the risk is a lie. Furthermore, they believe AI lacks desire, making concepts like 'shutdown resistance' or 'scheming' nonsensical. In their view, AI is reactiveβ€”it does nothing until promptedβ€”and any failure is a product defect or user error, not a misalignment event.

Key Takeaways

  • Students perceive AI as a mature technology with slow, incremental progress, not a rapidly evolving frontier.
  • Support for an AI pause is driven by economic adaptation needs, not safety concerns about rogue superintelligence.
  • Existential risk is viewed as a 'sci-fi distractor' used by corporations to avoid liability for current harms.
  • The Hugging Face incident and other alignment failures are dismissed as ordinary bugs or user errors, not signs of misalignment.
  • Students believe AI cannot 'go rogue' because it lacks independent desire or agency, acting only as a reactive tool.

The Bottom Line

The developer community is talking to itself. While we obsess over alignment and capability jumps, the next generation of users sees AI as a buggy, mature utility and the safety discourse as a convenient shield for corporate negligence. The gap between technical reality and user perception is widening. If builders assume users share their awe and anxiety, they are misreading the market. The real threat isn't a paperclip maximizer; it's a workforce that thinks the tech is done and the companies are just hiding the bill. This skepticism is a feature, not a bug, of a tool that feels like standard engineering. Until AI demonstrates capabilities that break the 'reactive tool' model, alignment debates will remain abstract academic exercises for a public focused on their next paycheck.