KAIST Develops AI That Avoids Hallucinating Even at Night or in Smoke (IMAGE)

The Korea Advanced Institute of Science and Technology (KAIST)

KAIST Develops AI That Avoids Hallucinating Even at Night or in Smoke

Caption

Figure 2. Cross-modal hallucinations and their mitigation through Modality-Adaptive Decoding (MAD). Influenced by the boat shown in the video, the base model hallucinates non-existent visual content and audio events—including the splash of a fish jumping out of the water—shown in red and blue text. MAD adaptively suppresses cross-modal interference, producing an accurate description grounded in the actual visual and audio content, shown in green text. (Courtesy of KAIST)

Credit

KAIST

Usage Restrictions

No

License

Licensed content

Disclaimer: AAAS and EurekAlert! are not responsible for the accuracy of news releases posted to EurekAlert! by contributing institutions or for the use of any information through the EurekAlert system.