Most of the open internet is low-quality — full of duplicated text, shallow SEO copy, spam, contradictions, and clickbait. Large AI models must ingest and compress enormous volumes of this noise just to extract a usable signal. They become good at pattern recall, but their reasoning degrades because the signal-to-noise ratio is so low. This wastes compute, energy, and model capacity. If AI systems were trained and grounded on high-quality, well-structured content, they could be dramatically smaller, cheaper to run, and far more widely deployable — even on local or edge devices. To realize their full potential, we need to give AI models high quality content.
via Andrej Karpathy, a prominent AI researcher, https://x.com/slow_developer/status/1979676364157653238
", "Notes": "" }