@DerriereLaLune I find AI is similar to other technologies, such as the automobile, in this regard: situationally very useful, but not universally so. (The article "AI as Normal Technology" https://knightcolumbia.org/content/ai-as-normal-technology is one elaboration of this point of view.) Technology is all about tradeoffs; it is almost impossible to avoid conceding some valuable good when using technology to obtain another valuable good, but the best one can do is be mindful both of what one is gaining and what one is giving up when using any technological tool, and explore ways to capture more benefits while simultaneously mitigating or reducing costs.

For instance, the arXiv summaries I was able to get my agent to generate are certainly extremely convenient compared to manually reading all the submissions (which has become quite challenging in recent months due both to increased volume of arXiv activity, and decreased time available on my end), but come at the cost of serendipity - finding an unexpectedly interesting paper that would not have made it to my summary list due to its offbeat nature or lack of evident match with my previously declared interests. I am thinking of directing my agent to try to create some "artificial serendipity" by throwing in a few additional wildcard items to the summary lists, but this is a partial mitigation of the issue at best. (EDIT: a screenshot of the experimental version of this feature is attached.)