AI overview contains obvious mistakes

[DBraun]

I read the AI overview for my own paper (https://www.alphaxiv.org/abs/2608.18226) and found obvious mistakes. For example, it says that "SLAP" means Synthesizer Audio-Parameter Shared Embeddings. SLAP actually means Siamese Language-Audio Pretraining. Also, there is no disclosure of which AI model wrote the overview, so readers don't know how to calibrate their understanding of the overview's accuracy.

So I suggest:

1. Use a better model for writing the overviews.
2. Disclose the model in the overview and the date the overview was created.

A related idea: in the AI Detection panel, share the same results across users so that users aren't redundantly running the AI detector. Also show the date and model.

[lino-levan]

1. Super sorry about this! We're actively working on a v2 version which should be significantly better, hoping to have it rolled out by the end of this week. We don't currently store which version of the overview generator we used so we're gonna start doing that to allow users to request an "upgrade" to the newer overview versions. We're also going to allow authors to edit the overviews which should help as well.
2. Totally agreee
3. AI Detection is shared across users, though it must be requested by at least one to run (it's pretty expensive for us since papers tend to be very long). Right now we use Pangram v3, but we're working with their team to upgrade to Pangram v4.

Hope that makes sense! Let me know if you have any other feedback.