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.
Why does Perplexity sell “end-to-end projects” as orchestration of 19 models, not one giant model?
Orchestrating many models turns a project into routed subtasks (reasoning/research/long-context), so capability scales via specialization+parallelism, not model size.
2026-02-24Why did xAI’s case fail even though two ex‑employees admitted stealing trade secrets?
Trade‑secret claims fail structurally when proof stops at employee theft: liability hinges on linking the new employer’s direction or use, not mere hiring flows.
2026-02-23What does “MLS-based E2EE for RCS” reveal about who must coordinate before secure iPhone–Android messaging ships?
MLS E2EE for RCS sits in the GSMA Universal Profile carrier layer, so Apple/Google can’t ship it alone—carrier profile rollout gates iPhone–Android encryption.
2026-02-22Which political bargain turns “equitable, human‑centric AI” into dairy yields, fraud vigilance, and cultural preservation demos?
Summit declarations operationalize “human‑centric AI” by binding legitimacy to measurable public‑goods pilots (agri yield, fraud safety, heritage digitization) rather than abstract ethics.
2026-02-21Where did Microsoft’s sensitivity labels and DLP policies break down—permission checks or Copilot’s summarization pipeline?
Label/DLP enforcement failed because Copilot’s summarization runs in a separate content-processing path where email content enters before label/policy gates, so a bug bypassed them.
2026-02-20How does “sovereign compute” coexist with OpenAI, Microsoft, and Nvidia anchoring India’s AI expansion?
“Sovereign compute” is enforced by locating data centers and model serving in-country; foreign firms supply GPUs/tools while governance stays with Indian partners.
2026-02-19What converts Apple’s iCloud CSAM controversy into a consumer-protection lawsuit demanding stronger detection controls?
Consumer-protection framing hinges on a “known misuse + inadequate controls” architecture: low CSAM reports become evidence Apple omitted detection/reporting as a safety feature.
2026-02-18How did Microsoft 365 Copilot bypass data-loss prevention specifically for Sent Items and Drafts?
DLP protected Inbox access but not Sent/Drafts because Copilot’s folder-scoped retrieval path skipped label/policy checks, so confidential mail entered summaries.
2026-02-17Why would a model company buy a 13-person serverless startup to claim “sovereign” European AI infrastructure?
“Sovereign AI” depends on owning the deploy/runtime layer: serverless GPU scheduling with sub‑second autoscale lets a model maker run EU‑hosted inference without US clouds.
2026-02-16Why is Anthropic treating India’s multilingual training data quality as a go-to-market lever, not research hygiene?
Multilingual data quality becomes a GTM lever when enterprise distribution runs through local integrators; language coverage turns into a sellable integration edge.