AI: Accelerated Incompetence

Original Article ↗ Hacker News Discussion ↗

AI Slop, Quality, and Discoverability

  • Many see “AI slop” as just faster, cheaper slop that would have existed anyway, but with much higher volume and lower barrier to entry, echoing what DAWs did to electronic music and app stores did to software discoverability.
  • Some argue the absolute volume of good output has increased, but the ratio of good to bad has worsened, making quality harder to find.

Productivity, Legacy Code, and Tech Debt

  • Claims that great engineers can get 4× productivity are heavily disputed. LLMs help most with loosely coupled, brownfield code; tightly coupled legacy systems remain hard for them to modify safely.
  • Several posters stress that prompting often clashes with established workflows and increases context switching; for many, “AI as mandatory for everything” is experienced as a net drag.
  • Others say AI is powerful for one-off utilities, CSV/ETL glue, visualization snippets, and “super-autocomplete” in typed languages, but fails on large, safety- or money-critical systems.

Cleanup Work vs Bubble Popping

  • One camp expects years of high-value cleanup and redesign after AI-generated messes, analogized to the post‑outsourcing correction era; another thinks this is wishful thinking and that companies will just stack more AI on top of AI.
  • A darker view: AI is a hype bubble like past fads; when it underdelivers, investment and jobs across tech will be hit, not generate a golden age of craftsman maintainers.

Concepts, Reasoning, and Complexity

  • Strong disagreement over whether LLMs can “work at a conceptual level” or hold program theory.
  • Critics argue LLMs are sophisticated token mimics lacking true concepts, counterfactual reasoning, or entropy-reducing design ability; any appearance of understanding is “cheating” via training data.
  • Defenders point to embeddings, internal concept activations, and practical use in refactoring or simplification when explicitly asked, claiming differences are of degree, not kind.

Skill Atrophy, Education, and Work Ethic

  • Multiple people report personal skill regression and “blanking out” after over-relying on AI, likening it to calculators and GPS degrading mental arithmetic and navigation.
  • Academia is cited as a domain already transformed: prior assessment and remote-teaching norms are breaking under ubiquitous LLM use.
  • There is concern that AI will degrade everyone’s work ethic and thinking, not just low performers, while management chases quantity over quality.

Analogies and Broader Framing

  • 3D printing is a recurring analogy: genuinely useful, even transformative in niches, but nowhere near “replacing all manufacturing.” Many think LLMs will follow a similar arc.
  • Several conclude AI is a powerful accelerant: it makes both good and bad engineering easier and faster, so institutional incentives and human judgment remain the real leverage points.