Baidu, Xiaomi CEOs warn of growing gap between AI hype and returns
Executives at Baidu and Xiaomi warning about a growing gap between AI hype and financial returns prompt wide skepticism over current business models for large-scale AI. Commenters generally agree that language models are powerful tools—especially for coding and personal “micro‑software”—but argue that integration into real organizations is slow, ROI is often unclear, and much investment looks bubble-like. Many foresee AI surviving as a valuable commodity technology while today’s frontier-model economics, cloud-heavy architectures, and “vibecoded” slop apps face correction, all amid geopolitical tensions and worries about labor, creativity, and enshittified software.
Perceived Hype vs. Real-World Returns
- Many see a growing gap between frontier-lab promises (10x productivity, “techno-god”) and measurable business impact.
- Some report users feel more productive but metrics don’t improve.
- Others argue this is just normal hype-cycle timing; tech is real, but diffusion and org change take years.
Adoption Frictions Inside Organizations
- Biggest blockers cited: security reviews, connectors, infra, governance, and plain “too much other work.”
- Integrating AI into existing workflows and processes is slow and politically fraught.
- Some pilots show strong results (log analysis, code review, risk assessment), but turning PoCs into scaled systems is hard.
Where AI Appears Useful
- Strongest consensus: coding assistance and small internal tools. Solo devs report building niche apps at unprecedented speed.
- Other wins: OCR quality boosts, better enterprise search in some orgs, L1 support deflection, HR self-service, legal drafting, back-office form processing.
Concerns About Quality, “Slop,” and Reliability
- Many complain about “vibecoded” software: fragile, unmaintainable, hallucination-prone.
- Visible outages and poor systems are sometimes blamed (fairly or not) on AI and post-layoff culture.
- Creative domains see floods of low-effort content (books, music, art, restaurant menus), raising worries about taste, originality, and signal-to-noise.
Economics, Business Models, and Moats
- Skepticism that massive spend on frontier models will pay off before being commoditized by open weights and distillation.
- Some argue cheap local or open models will dominate once hardware normalizes; big US labs could face an “Iridium-style” wipeout.
- Others note cloud providers are currently making money on AI, but worry about sustaining growth and downstream customer ROI.
Labor, Capitalism, and Organizational Change
- Managers in the thread talk openly about replacing or not hiring humans where an agent suffices.
- Counterpoint: AI often automates fractions of many roles, making direct headcount cuts politically and socially difficult.
- Broader critique targets “financialized capitalism”: AI framed as a labor-elimination machine and rent-seeking over “means of production.”
Open vs Closed and Geopolitics
- Debate over Chinese vs US models: some see open-weight releases (including from China and big US clouds) as economic warfare; others as normal commodification strategy.
- National security rules, export controls, and trust in foreign models are seen as shaping which providers enterprises can actually use.