Ask HN: Is anyone else bearish on OpenAI?

Hacker News Discussion ↗

Skepticism about OpenAI’s long‑term prospects is colliding with reports of dramatic day‑to‑day productivity gains from tools like ChatGPT. Commenters compare the current AI boom to the crypto hype cycle, questioning AGI promises, business moats, and high compute costs, while others say LLMs have already replaced much of their web search, debugging, scripting, and writing workflows despite hallucination risks. Many expect OpenAI to face serious competition from open‑source models and tech giants, but see large language models in general as a durable shift in how people interact with computers.

Perceived Practical Value of ChatGPT/LLMs

  • Many commenters say GPT-4 (and Bing Chat based on it) has become a daily tool, often replacing Google/Stack Overflow for “how do I do X in Y” questions.
  • Reported gains: faster boilerplate coding, SQL/jq/shell one‑liners, Terraform/Terraform refactors, debugging obscure errors, writing tests, refactoring legacy configs, small scripts, data cleanup, copywriting, job descriptions, emails, reports, translations, survey summarization, and document → Markdown conversions.
  • Some say it materially changes SRE/devops workflows (kernel/network deep dives, error interpretation, config issues), and that they’d pay much more than $20/month.
  • Several non‑dev uses: education/tutoring, language learning, rewriting letters, legal‑style correspondence drafts, UX/management brainstorming, tax/letter reviews, and content proofreading.

Limits, Hallucinations, and Trust

  • Many report “confidently wrong” answers, made‑up APIs, libraries, academic papers, and code that doesn’t type‑check or run.
  • Hallucinations are seen as acceptable by some if you verify like you would with Stack Overflow; others find the verification overhead negates any benefit.
  • GPT‑4 is widely described as “a different level” from 3.5/Claude Instant, yet still error‑prone and occasionally nonsensical.
  • There is concern that it sounds authoritative, encourages over‑trust, and is unsuitable as a sole source for medical/legal/scientific facts.

Who Benefits and For What Tasks

  • Strongest wins are in “complicated but shallow” tasks: wiring APIs, infrastructure as code, CSV/log analysis, quick code patterns, summarization, language/style work.
  • It works poorly for deep theoretical CS/math (NP reductions, formal semantics, advanced algorithms) and niche domains with sparse training data.
  • Several note it is most useful when you already know enough to check its work; dangerous when you don’t.

Comparisons to Crypto and Hype Concerns

  • Some see echoes of crypto: massive hype, wrapper startups, VCs piling in, and vague promises of AGI.
  • Others argue the key difference is clear, immediate utility to millions, unlike crypto’s limited real‑world use.
  • A subset is “meh”: they tried GPT, found it generic or wrong, and feel Google/docs + expertise are still better.

OpenAI’s Business Prospects and Moat

  • Views range from “easily $100B+ company / next Google‑tier juggernaut” to skepticism about margins, GPU costs, and lack of moat as open/open‑source models improve.
  • Some expect regulatory capture and a pseudo‑monopoly; others think Meta, Google, Microsoft, or on‑device models will erode OpenAI’s lead.
  • Concern that many “OpenAI wrappers” will be crushed when OpenAI productizes the best ideas itself.

On‑Device, Privacy, and Open Source

  • A privacy‑minded minority wants near‑GPT‑4 models running offline on personal devices; current local models are seen as too slow/weak or too hard to operate.
  • Many are bullish on smaller open‑source models at the edge as a major long‑term threat to cloud‑only offerings.

AGI, Trajectory, and Society

  • Many don’t think current LLMs are AGI or sufficient to reach it; some see them as one component among future methods.
  • Others argue that the ability to compress world knowledge and perform few‑shot learning makes AGI feel like “an implementation detail,” though timing is unclear.
  • Expected near‑term impacts: large productivity boosts, displacement of lower‑tier white‑collar work (support, basic writing, junior coding), and significant changes in education and “how we interact with computers,” even without AGI.