A claim cited by a paper that cited a paper can look stronger than the evidence ever was. AI could turn that into a feedback loop. Here is the research on the problem, and four defenses I would build into ABE.

In my last post I argued that the browser will soon build pages instead of fetching them. Here is how I would build that browser so it stays open: an open standard first, your choice of AI model, and shared rules that let any agent plug in safely.