Satya Nadella: "We need to find something useful for AI"

AI Hype vs. Real Utility

• Many see Nadella’s remark as a rebranded “you’re using it wrong,” shifting blame to customers for not extracting enough value to justify massive AI spending.
• Others interpret his actual wording (“do something useful that changes outcomes”) as a rare big‑tech emphasis on human benefit, but argue the messaging is muddied by hype and over‑investment.
• Several compare the moment to the dot‑com bubble or a gold rush: lots of “shovel factories” (infrastructure, chips) built ahead of proven demand.

Productivity Gains: Incremental vs. Transformative

• Some commenters report substantial personal gains: cutting legal/administrative costs, rapidly prototyping small apps, doing ad‑hoc data analysis, and treating LLMs as powerful “power tools.”
• Others (including non‑tech users) mostly see AI as a toy for silly questions, or an only‑slightly‑better search/StackOverflow.
• A recurring theme: improvements are often incremental, not the massive economy‑wide productivity leap implied by the investment levels.

Costs, Energy, and the “Social Permission” Framing

• Skepticism that “social permission” around energy will be the binding constraint; many think investor patience and operating costs will bite first.
• Concerns that AI is already driving up prices for GPUs, RAM, flash, and now electricity, with resistance to power‑hungry data centers in some regions.
• Some predict cheap AI will “enshittify” into ad‑driven, nerfed services, with high‑quality models reserved for those paying full cost or running their own hardware.

Hallucinations, Trust, and Use Cases

• Numerous stories of hallucinated APIs, OS settings, URLs, financial transactions, and medical explanations; people note the extra time spent verifying.
• Others argue hallucination is overstated in modern tools, especially when they actually use web search and cite sources.
• There’s debate over whether LLMs meaningfully “amplify cognition” versus masking incompetence, dulling skills, or merely laundering existing information.

Domains and Misaligned Promises

• Claimed strengths include coding assistance, documentation drafting, personal data querying, and search summarization; skeptics question scalability and ROI for trillion‑dollar bets.
• Use in healthcare is seen as over‑sold: realistic benefits in admin and explanation, but not miraculous diagnosis.
• Spam, porn, and sophisticated manipulation are widely viewed as “surefire” use cases, raising social and ethical concerns.

Microsoft’s Strategy and Copilot

• Many find the proliferation of Copilot branding (Office, Windows, everywhere) underwhelming or gimmicky, seeing a mismatch between rhetoric about “useful AI” and actual shipped products.