Adapting Large Language Models for Long-Context Reasoning in Challenging Domains

EURAXESS hosting organisation · Plusieurs pays européens

Modern Large Language Models (LLMs) can process very long inputs, with context windows extending to hundreds of thousands of tokens. However, they often struggle to integrate information across the entire context (for example, when resolving causal chains or tracking the state of entities mentioned throughout a document). The selected candidate will be responsible for researching techniques for improving long-context reasoning, spanning from texts with an explicit structure, such as encyclopedias and technical documentation, to narrative texts, where the implicit structure requires the model to reason over the entire document rather than relying on explicit references, and potentially to leverage information stored in external knowledge bases or document repositories.

Who can apply

Criteria stated by the funder on the official source.

Funding and what it does not cover

Funding details were taken from the official programme page text. Confirm the current call before applying.