rahulb0802

Since you wrote the detector using a particular style guide, are there other indicators of AI writing you are still uncovering and adding to this? Or do the ones mentioned mostly encompass everything?

What does the actual path to deployment look like? Do I connect a repo, copy paste code, or something else?

How did you assemble your training dataset, especially since you mention some of the considerations with different training mixes in your controlled eval?

I really like the clustering aspect of this. It would be cool if there was a way to articulate some sort of relationship between two adjacent or nearby papers in this tool.

I like the approach; avoiding a model call just to summarize old context while keeping the cache warm is a nice property.

When it folds away an older chunk of the conversation, does it cause problems if the agent was in the middle of doing something (some multi-turn action) right when that chunk gets folded? Or is it safe to fold at any point regardless of what the agent was doing at that time?

This is cool! One question: is the format documented anywhere, or a stable enough schema that someone could write, for example, a script that pulls last week's jots into a review? Having an implied contract that the format won't change as opposed to a JSON that just happens to be inspectable today would be more compelling I feel.