"Don't You Just Upload It to ChatGPT?"

Original Article ↗ Hacker News Discussion ↗

AI capabilities and trajectory

  • Many argue translation and coding are already at or near a “tipping point,” with LLMs often faster, cheaper, and “good enough” for many tasks.
  • Others stress current systems are unreliable, prone to factual and contextual errors, and fundamentally not oracles.
  • There’s debate over extrapolation: some see rapid, compounding progress leading to widespread replacement; skeptics warn against assuming linear or infinite improvement or ignoring local maxima.

Translation quality and use cases

  • Broad agreement that AI is excellent for informal or personal translation: web pages, documents for personal understanding, quick gist.
  • Strong pushback that professional, high‑stakes, or artistic translation (literature, poetry, nuanced dialogue, legal/technical docs, UI texts) still requires humans for tone, cultural context, consistency, and “voice.”
  • Some claim AI has raised the floor relative to bad human translators; others say average quality has dropped as cheap machine output proliferates.
  • Examples show both human disasters and machine disasters; the consensus is that context and stakes matter.

Economic and labor impacts

  • Translators report shrinking markets and downward pressure on rates; similar fears surface for software developers, call centers, voice actors, etc.
  • Several expect many roles to shift from production to auditing/oversight of AI output, though pay and demand may shrink.
  • Others emphasize that society chooses how to deploy AI; profit motives, not capability alone, drive job loss and “enshittification.”

Experts, trust, and AI “slop”

  • Repeated theme: people trust AI for other fields but see its flaws in their own, an AI-flavored Gell‑Mann Amnesia/Dunning–Kruger pattern.
  • Some warn about overconfidence: AI output can look authoritative while being subtly wrong, increasing review burden for experts.
  • A minority report huge productivity gains by orchestrating multiple agents and treating code as almost disposable, while others working on higher‑risk systems see far less benefit.

Quality vs cost and “good enough”

  • Many expect organizations to accept lower quality if it’s much cheaper and faster, especially where users can’t easily judge quality.
  • Others insist there will remain niches—high‑stakes, aesthetic, or regulatory—where certified human work is still demanded.