@ale_suglia : LLMs are great but they are brittle to minimal prompt perturbations (e.g., typos, indentation, ...). Q: How do we create truly multimodal foundation models? A: Do as we humans do: text as visual perception! Enter PIXAR, our work at #ACL25024NLP! arxiv.org/abs/2401.03321 • TwiCopy

Alessandro Suglia

@ale_suglia

+ Follow

Assistant Professor @HeriotWattUni/@NRobotarium; Ex Head of Visual Dialogue at @helloalana; PhD @EDINRobotics; Ex Research Intern @MetaAI and @AmazonScience.

ID: 1213616077

linkhttps://alesuglia.github.io calendar_today23-02-2013 20:22:02

2,2K Tweet

1,1K Takipçi

1,1K Takip Edilen

Alessandro Suglia

@ale_suglia

2 months ago

LLMs are great but they are brittle to minimal prompt perturbations (e.g., typos, indentation, ...). Q: How do we create truly multimodal foundation models? A: Do as we humans do: text as visual perception! Enter PIXAR, our work at #ACL2024NLP! arxiv.org/abs/2401.03321