Don't make me talk to your chatbot

Original Article ↗ Hacker News Discussion ↗

Growing use of AI chatbots—in customer support, corporate phone systems, and even code review—has triggered frustration from people who feel fobbed off by unhelpful automation and formulaic “AI slop.” Commenters weigh the economics that push companies toward bots (support calls are expensive, margins are thin) against the rudeness and inefficacy of blocking access to competent humans, especially for complex or edge‑case problems. Others worry more broadly that overreliance on LLM‑generated text, from blog posts to pull request descriptions, erodes trust and signal quality online unless humans do the hard work of thinking, providing real context, and clearly owning what they publish.

Meta: Many commenters didn’t read the article

  • Large portion of the thread treats the title as being about customer-support chatbots.
  • Multiple people point out the article is actually about humans offloading their writing/thinking to LLMs and then making others read that output.
  • Some argue HN behaves like other social media: reacting to the headline, not the content.

Customer-support chatbots: experiences and tradeoffs

  • Some users like chatbots as first-line support: instant responses, quick refunds, or painless price negotiations (e.g., deliveries, subscriptions).
  • Others say chatbots rarely solve real problems, serve mostly to stonewall, or funnel users into dead ends (broken callback systems, limited options, repeated data entry).
  • A popular “good” pattern: bot collects structured info, then hands off to a human (“smart answering machine”).
  • Frustration focuses on lack of reliable human fallback, especially for banks, ISPs, government portals, and complex billing issues.

Economics and ethics of support

  • One perspective: human support is very expensive (training, churn, facilities, full burdened costs); trivial calls like password resets or “power cycle your router” dominate volumes.
  • Counterpoints:
    • Big firms helped create these problems with confusing UX, fragile products, and opaque flows (esp. identity and passwords).
    • Customers have already paid; “free support” is just bundled support.
    • High-value calls (real bugs, deep technical issues) are rare but important; current triage systems make them too hard to report.
  • Debate over charging for support with refunds if it’s the company’s fault; concerns about perverse incentives to deny responsibility.

“AI slop” in writing, PRs, and discussion

  • Strong dislike for generic, verbose LLM-generated prose in PR descriptions, blog posts, Slack, LinkedIn, etc.; seen as low-signal, formulaic, and often wrong on details and intent.
  • Several argue:
    • If you had to think hard enough to prompt the model well, you could have just written the thing.
    • Readers care about your reasoning and relationship to the facts, not a synthesized average of internet text.
    • LLMs act as “misunderstanding amplifiers” when given fuzzy internal concepts or jargon.
  • Others see value in LLMs as:
    • Tooling to expose complex systems via natural language interfaces.
    • Grammar/spell-check and expansion of terse points into accessible prose.
    • Triage aids that surface relevant docs or APIs, as long as humans provide a concise, honest “anchor” summary.

Broader worries about AI content

  • Concern that AI-generated “slop” will further drown already noisy internet content, making high-signal material harder to find.
  • Calls for emerging etiquette: don’t use agents as your voice in genuine human exchanges, and don’t force others to “talk to your chatbot” when they came to talk to you.