Question archive
Each day, we pose a question inspired by the daily news brief and answer it using our database of indexed AI/ML articles. Browse and search past questions below.
Where does a “photo suggestions” feature cross from convenience into unlawful device-level data access?
“Convenience” becomes unlawful when device photo access is bundled into broad, default-on permissions; regulators require a separate, specific opt‑in gate per data type.
2026-11-13How can Cerebras post adjusted profit while its GAAP net loss is widening?
“Adjusted profit” can coexist with a widening GAAP loss because non‑GAAP reporting structurally strips out GAAP-required cost buckets (e.g., SBC, amortization), recasting economics.
2026-11-12How does invoking California’s unfair competition law let nonprofits seek an AI-development halt without an AI-specific regulator?
California’s Unfair Competition Law functions as a general-purpose enforcement hook: by framing AI training as a public-welfare risk, plaintiffs can seek injunctive pauses without an AI regulator.
2026-11-11Why does IBM highlight “10,000 quantum operations” on a 520‑qubit chip, not just qubit count?
Usable quantum compute scales with how many gates run before errors accumulate, so “10,000 ops” measures circuit depth/fidelity, not raw qubit inventory.
2026-11-10Why is OpenAI letting businesses “claim” ChatGPT profiles instead of letting agents infer product facts from the web?
Claimed profiles shift truth maintenance from brittle web-scraping to an owner-managed “source of record,” aligning accuracy with update control and analytics.
2026-11-09How does an EU-wide “ten dark patterns” ban turn UX choices into a fines-and-enforcement compliance surface?
By enumerating banned UX tactics, the EU makes interface flows auditable like product claims, so enforcement targets specific interaction patterns, not just outcomes.
2026-11-08Why is an AI model company paying $3.4B for 1.7GW of gas generators instead of just buying more GPUs?
AI scale is power-connection limited: grid interconnect queues and long-lead transformers, not GPUs, set the pace, so firms buy on-site generation to unblock capacity.
2026-11-07Why would Chinese intelligence target Nvidia’s export data—when US chip export rules are already public?
Even with public export rules, enforcement hinges on opaque distributor logs and buyer networks; export data maps the weakest nodes—shell firms, drop‑ships, smuggling routes.
2026-11-06How does buying Qualcomm shares advance Alphabet’s data-center AI push without building a single chip?
A minority equity stake buys “influence + alignment” in a supplier’s roadmap and partnerships, letting a hyperscaler shape data‑center AI hardware via ecosystem leverage, not fabs.
2026-11-05Where does Claude Personal Finance’s real risk surface if Plaid links are read-only?
Read-only Plaid shifts risk from fund-movement to the data plane: the exchange, storage, and tool-integration paths become the main exploit surface.