I love LLMs, I hate hype

Original Article ↗ Hacker News Discussion ↗

LLMs are seen by many here as powerful new programming and creative tools, but frustration is growing with the surrounding hype, doomerism, and promises of instant fortunes or imminent obsolescence of all knowledge work. Commenters argue that current systems are “just software” built on decades of computing progress, useful but not magical, and that much of the value they create will likely be captured by hardware makers and practical downstream applications rather than today’s frontier labs. Concerns center on long‑term costs, vendor lock‑in, and social impacts—from job displacement to “brain rot”—alongside optimism that local and open models, used thoughtfully, can deliver lasting benefits without requiring everyone to buy into Silicon Valley culture.

Hype, Fear, and “Perpetual Underclass” Narratives

  • Many dislike both positive and negative hype: doom about being “left behind” and utopian AGI talk are seen as manipulative and mostly marketing.
  • Some see little first‑hand of the extreme hype and more of a backlash/anti‑hype culture on forums and at work.
  • Others report constant exposure to conspiratorial or catastrophist views (on social media, local politics, etc.).

San Francisco and Tech Culture

  • Strong disagreement over bashing SF: some argue cost‑of‑living and governance justify criticism; others say blaming one metro is lazy and ignores its diversity.
  • Several note that moving to SF for AI/AGI networking is part of the hype machine.

Builders vs Merchants

  • Distinction drawn between “merchants” (middlemen platforms building moats on others’ work) and “craftsmen” who build and directly sell their own products.
  • Some argue anyone selling AI products is a merchant regardless of narrative.

LLMs as Tools and “Vibe Coding”

  • Many use LLMs heavily for coding, brainstorming, documentation, and “vibe‑coding” small personal tools.
  • Gains: fewer stuck moments, easier prototyping, domain experts building complex tools without hiring devs.
  • Limits: models still produce “slop” if unsupervised; real productivity constrained by design, scoping, maintenance, and integration, not raw code generation.

Cognitive Effects

  • Some feel LLMs are “poison for the brain,” encouraging offloading thinking and atrophying skills; others compare this to historical fears about writing and calculators.
  • Several report using LLMs to tackle harder problems than before, arguing challenge level, not tool, determines mental impact.

Art, Authenticity, and Meaning

  • Debate over whether AI can create “art” or only “artifacts.”
  • One side ties art to human emotional experience and creator–audience social contracts; others note much commercial art optimizes for induced emotion and see no principled bar to AI.

Open Source, Forking, and Maintenance

  • LLMs make forking and customizing OSS vastly easier, enabling “have it your way” personal software.
  • Concerns: fragmentation, harder upstream collaboration, and future “maintenance hell,” though some argue many tools will reach “good enough” and be left alone.

Costs, Business Models, and Commoditization

  • Disagreement on whether current frontier models are subsidized or already profitable on API pricing.
  • Subscription plans are widely seen as loss‑leaders; real money is in enterprise/API usage.
  • Skepticism that frontier labs can sustain sky‑high valuations if cheaper open or “good enough” models proliferate.
  • Some predict AI will resemble other infrastructure (airlines, PCs): huge social value but limited value capture by any one lab; hardware vendors may benefit most.

Local vs Frontier Models and Future Access

  • Many are gradually shifting toward local and open models for cost, control, and censorship reasons, using frontier models only when necessary.
  • Expectation (though not guaranteed) that model quality per FLOP will improve enough for near‑frontier capability on consumer hardware; disagreement on timelines and physical limits.

Singularity, ASI, and Long‑Term Outcomes

  • Split between those who see superintelligence as radically disruptive and dangerous and those who think “flash of light” scenarios are overblown.
  • Some worry less about runaway AI itself and more about aligning it with existing shareholder‑driven, extractive economic systems.