AI slop floods science, Kathmandu Post warns of research risks

This column captures a growing anxiety inside academia: as generative models get better at producing “good enough” text, they start to overwhelm scientific channels with syntactically polished but intellectually shallow work. If large portions of preprints, reviews and even grant applications are largely AI written, the signal-to-noise ratio in the knowledge pipeline drops. For anyone trying to track real progress toward AGI, that makes benchmarks, citation counts and even expert consensus much harder to interpret. ([kathmandupost.com](https://kathmandupost.com/columns/2025/12/21/ai-slop-and-science))

It also underlines a paradox of the race to AGI. The same tools that accelerate ideation and coding can erode the very epistemic infrastructure—peer review, careful writing, reproducible argument—that advanced AI research depends on. If academic incentives stay quantity-driven, AI assistance will amplify existing pathologies: salami-sliced papers, overstated claims, and shallow literature engagement, now scaled by LLMs. That forces serious labs, investors and regulators to invest more in metadata, provenance and automated review just to keep track of what is real.

Longer term, we should expect a bifurcation: elite venues and funders tighten rules around AI use, while lower-tier outlets drown in synthetic content. For Race to AGI readers, the message is clear: you’ll need more rigorous filters—technical, social and institutional—to separate genuine breakthroughs from AI-generated academic slop.