From curiosity to contribution
A recent paper in the field, “LLMs Get Lost in Multi-Turn Conversation”, helped refine Dayyán’s research focus.
“It showed that language models often latch onto an incorrect assumption early in a conversation and fail to correct themselves, even with new information. This resonated with my own experience using ChatGPT and similar models. It’s a problem of reliability. Models have the capacity to solve a problem, but cannot apply it in the most natural setting for humans: conversation.”
For Dayyán, improving how models compose language could be key to solving that.
“A model that understands how pieces of language fit together should be more flexible and better at fixing its own mistakes. It’s a step towards making AI systems that are more robust, interpretable and trustworthy.”
Opening doors through NextGen
For Dayyán, the G-Research Scholarship provides more than funding – it offers mentorship, collaboration and a connection between academic and industry research.
“I’m most excited about how G-Research bridges the gap between academia and industry. Being part of a community that applies cutting-edge research to real problems is exactly what I’m looking for. Learning from both researchers and practitioners will be an incredible opportunity.”
