Pew Study: 35% of Post-ChatGPT Webpages Show Signs of AI Authorship

Skynet Chance (+0.02%): Mass AI-generated web content degrades the provenance and quality of the primary training corpus for future models, creating recursive-training and hard-to-audit data contamination risks that complicate alignment and oversight. The effect is diffuse and indirect rather than a direct loss-of-control mechanism.

Skynet Date (+0 days): Contaminated training data may modestly slow the reliable scaling of frontier capabilities, marginally pushing out rather than accelerating risk timelines. The influence on pace is minor and speculative.

AGI Progress (0%): The finding measures deployment scale and diffusion of existing generative text tools rather than any new capability, though it evidences that current models are good enough to pass as human across a large share of the open web. It shrinks the pool of clean human-authored data, a mild headwind for capability gains.

AGI Date (+0 days): Shrinking availability of verified human-written text is a data-supply constraint that could slightly decelerate progress that depends on scaling high-quality corpora. The magnitude is small relative to synthetic-data and compute-driven advances.