Ben Lorica: AI's Data Problem Moved Downstream — Usability, Not Scarcity, Is the Bottleneck
bigdata · x · 2026-09-30
Ben Lorica argues in Gradient Flow that AI's data problem has shifted from finding information to making it usable:
- Robotics: Unlike LLMs' decades-accumulated internet corpora, robots lacked a comparable internet of physical experience — but one company trained on 1M+ hours of human video, and another converted 1,900 hours of first-person footage into 18,000+ hours of robot-format data. The key insight: every system needs substantial machinery to translate human experience into robot-usable form.
- Licensing fragmenting: An academic publisher enabled retrieval for AI research products while explicitly withholding training rights — "use for AI" is splitting into separate permissions for training, retrieval, and inference.
- Access infrastructure: Sites can now distinguish search/training/agent crawlers, and one closed-beta system supports HTTP 402 responses with an access price — the web is becoming a market with rules.
Related event: Ben Lorica: AI's Data Problem Shifts to Usability(2 posts)→
More from AGI Musings
- Looking back at old Reddit threads mocking AI capabilities hasn't aged well — bowl_cut53 · 2026-09-30
- "Sharp Right Turn": why AIs suddenly appearing aligned should be a warning, not a relief — ZeroStateReflex · 2026-09-30
- Switching personal AI agents means exposing your privacy, deepening Big Tech lock-in — oran_ge · 2026-09-30
- Daniel Litt: once models write well, human writing will mainly help you think — littmath · 2026-09-30
- Mathematician Daniel Litt: next few years may see more math text than the past 1000 years — littmath · 2026-09-30
- repligate says Anthropic staff repeatedly pressed him to soften public criticism — repligate · 2026-09-30