Training large language models on narrow tasks can lead to broad misalignment

Journal:
Nature
Published:
DOI:
10.1038/s41586-025-09937-5
Affiliations:
9
Authors:
9
Go to article
Institutions Authors Share
Truthful AI, United States of America (USA)
1.500000
 0.19
Center on Long-Term Risk, United Kingdom (UK)
1.000000
 0.13
Warsaw University of Technology (WUT), Poland
1.000000
 0.13
UCL, United Kingdom (UK)
1.000000
 0.13
University of Toronto (U of T), Canada
1.000000
 0.13
Department for Science, Innovation and Technology (DSIT), United Kingdom (UK)
1.000000
 0.13
Stanford University, United States of America (USA)
1.000000
 0.13
University of California, Berkeley (UC Berkeley), United States of America (USA)
0.500000
 0.06
Interested in Control Engineering, Mechatronics and Robotics?