Pinned
New paper!! Sci-rho (Science-Rhobustness)
How much of VLM "reasoning" survives when the same problem is redrawn? We explore robustness across semantically equivalent visual variants.
Paper: arxiv.org/pdf/2606.08034
Thread below!!
IkhlasulHanif
@IkhlasulHanif0
AI Safety, NLP PhD student @mbzuai, currently visiting @sutdsg
Prev: @univ_indonesia and JAIST
Only create X because AI Researchers are on X
