Direct discovery · Computer science · General science

ChatGPT struggles to attach real evidence to domain answers

ChatGPT struggles to attach real evidence to domain answers.

Summary

The arXiv paper tests ChatGPT on domain-specific knowledge questions while asking for supporting references. It reports that answers were correct or partly correct in about half of cases, but suggested references existed only 14% of the time and often did not support the attributed claims.

AI role

AI contributes to generating, ranking, predicting, proving, designing, or validating a concrete research output in computer science, general science.

Narrative role

This belongs in the timeline because it records a concrete AI-assisted scientific output rather than only a workflow, product, or adoption signal.

Caveat

The event should be read as a bounded research result unless later validation, independent replication, adoption, or field-level impact is tracked separately.