All things AI

Large language models (LLMs) can “hallucinate,” generating incorrect information, which poses a considerable risk in high-stakes fields like healthcare or finance. To address this, MIT researchers created SymGen, a system that makes LLM verification faster and more accurate. SymGen allows models to generate responses with precise citations by referencing specific data points, such as cells in a table, so human validators can verify claims efficiently. The system highlights text that needs further attention and reduces verification time by about 20%. This innovative tool is being enhanced to work with diverse data formats, aiming to improve LLM reliability in real-world applications.

From MIT News