Goldman Sachs says the return on investment for AI might be disappointing

Original Article ↗ Hacker News Discussion ↗

Overall Hype, ROI, and “AI Winter” Talk

  • Many see current AI spending as classic overbidding driven by hype and FOMO, so near‑term ROI is expected to be poor.
  • Several compare this to previous bubbles (dot‑com, tulips, South Sea, blockchain/NFTs), expecting a bust and possibly an “AI winter.”
  • Others argue this is a normal “trough of disillusionment” phase for a genuinely transformative technology, not a sign the whole thing is a fad.

Comparisons to Past Tech Waves

  • Recurrent analogy: replace “AI” with “internet,” “big data,” “ML,” etc., and the pattern of clueless corporate strategy looks identical.
  • Lessons cited: some firms will find winning strategies (Amazon/Netflix), many will waste money or die (Webvan/Blockbuster).
  • Disagreement over which analogy fits best: internet/transistors (foundational) vs. blockchain/NFTs (overhyped, limited real value).

Corporate AI Strategies, FOMO, and Branding

  • Many non‑tech and even tech firms are seen as rushing into “AI strategy” without understanding capabilities or limitations.
  • FOMO is reinforced by employees, candidates, and customers asking about AI; “AI‑driven” branding can boost sales even if features are thin.
  • Some advocate the “Apple approach”: wait, use AI pragmatically where it clearly helps, avoid panic pivots.
  • Others argue every organization at least owes itself a serious evaluation of LLMs, even if it ultimately passes.

Practical Use Cases and Reliability

  • Positive reports:
    • Call/meeting transcription and summarization at scale.
    • Turning dense regulations into checklists.
    • Soft‑data summarization in finance and some data‑science workflows.
    • Generating mock data and structuring unstructured text.
  • Negative reports:
    • Frequent hallucinations in article summaries and code generation.
    • Output requires careful verification; “impressive demo” but not robust in complex real systems.
  • Some note that many current uses automate low‑value or unnecessary tasks (press releases, filler content).

Cost, Energy, and Efficiency

  • Concern that AI is too expensive in GPUs and power for broad deployment; subsidies and investor money currently mask true costs.
  • Counterclaim: LLM costs have dropped by ~10–50x in a year, with more gains expected, making ROI increasingly favorable.
  • Debate over energy metrics: “brain is 10,000x more efficient” is challenged; others say effectiveness per dollar, not per watt, is what matters.
  • Nvidia’s valuation and what happens if GPU prices collapse is raised as a risk for index investors.

Labor, Society, and AGI

  • Some predict near‑term heavy automation in security operations centers, call centers, and customer support; others are skeptical.
  • Mixed expectations about whether AI mainly augments workers or replaces large swaths of labor; downstream demand and regulation remain unclear.
  • Speculation around AGI: alignment, whether it would “work for humans,” possibility of utopia vs. entrenched inequality; timelines widely disputed and largely labeled uncertain.

Investment Dynamics

  • View that large investors knowingly ride hype, then exit before retail and latecomers absorb losses.
  • Disagreement over passive index funds: some think they’re risky given concentration in AI winners; others defend broad indexing as rational.
  • Several note that, as with dot‑com, much value may eventually accrue, but not necessarily to the companies currently burning the most cash.