How Does OpenAI Survive?
Skeptics question whether OpenAI’s massive spending on large language models can ever be justified by its current revenue and use cases, arguing that generative AI lacks a clear “killer app” and may already be hitting a technical and economic plateau. Others counter that chatbots, coding assistants, translation, and call-center automation are already delivering substantial value, and that rapid model and hardware improvements could still unlock transformative productivity gains. The exchange highlights uncertainty over whether AI leaders like OpenAI will become enduring, highly profitable platforms akin to Google and Microsoft, or symbols of an overextended investment bubble in need of a breakthrough that may never arrive.
Mass‑market utility and “killer app”
- Strong disagreement over whether generative AI has real mass‑market value yet.
- Skeptics say main uses are spam, low‑effort content, and demos; no obvious “killer app” comparable to early web search or online shopping.
- Supporters cite: coding assistance, tutoring, translation, call‑center replacement, documentation, text editing, and non‑engineers building small apps with chatbots.
- Some argue the chatbot itself is already a billion‑dollar product; others say that’s modest relative to current valuations.
Labor, agents, and automation
- One camp expects AI to replace large shares of white‑collar work (and later blue‑collar via robotics).
- Others think current systems are too fuzzy and liability‑prone; many tasks are better done with deterministic software or rigid scripts.
- “Agents” (LLM‑driven task automation/RPA‑like systems) are widely discussed as a likely high‑value direction, but their present capabilities are seen as limited and tooling‑dependent.
Economics, profitability, and scale
- Many doubt current spending on GPUs and training can be justified by near‑term revenue; inference and training are both expensive, and competition plus open‑source models compress margins.
- Some compare OpenAI to Stripe or Google’s early days (valuable but not obvious); others counter that Stripe solved a visibly painful problem with a clear path to profit, unlike LLMs.
- A recurring worry: foundation models may become a low‑margin commodity; incumbents with cloud scale or OS models could capture most value.
Trajectory: exponential breakthrough vs plateau
- Optimists argue capabilities per dollar are improving rapidly, scaling laws still work, and multimodal models/robots will unlock new domains.
- Skeptics see signs of a sigmoid: GPT‑4’s age, incremental model updates, limited qualitatively new abilities, and data constraints (especially text).
- There is debate about whether further gains will require major architectural breakthroughs, better data, continual learning, or just more compute.
Meta‑discussion and uncertainty
- Some argue that if transformative progress doesn’t arrive, big AI labs face serious financial strain; others note deep-pocketed backers and hype‑driven market caps may sustain them longer than critics expect.
- Participants highlight that prior tech booms had both spectacular successes and failures; many expect a shakeout where most AI startups die but some large players endure.