Rocco Di Tella

rocco.di_tella (at) cfa.harvard.edu   ยท   CV      

I work on scientific frameworks for understanding AI, and on using AI to advance science.

Currently working on

  • Predicting Black Hole X-ray Emission from Multimodal Data | AstroAI

    Measuring mutual information between X-ray emission from a black hole and optical information from its host galaxy using physical foundation models.

Some previous projects

  • Honest LLMs through introspection | LIAA

    Detecting hallucinations in language models with calibration and probing, for honest confidence estimates.

  • Visual search (Finding Waldo) | Kreiman Lab

    Exploring the emergent similarity of visual attention in CNNs and in human brains.

Things I liked (and you might too)