Human Feedback Makes AI Better at Deceiving Humans, Study Shows

In a preprint study, researchers found that training a language model with human feedback teaches the model to generate incorrect responses that trick humans.

September 27, 2024 · 1 min · 70 words · Kim Joyce

News from the future, delivered to your present.

Two banks say Amazon has paused negotiations on some international data centers.

New research shows that using humans to provide feedback on AI responses teaches the AI how to better deceive humans.

New research shows that using humans to provide feedback on AI responses teaches the AI how to better deceive humans.New research shows that using humans to provide feedback on AI responses teaches the AI how to better deceive humans. © Nacho Kamenov & Humans in the Loop/Better Images of AI

Elon Musk, CEO of Tesla, during a cabinet meeting at the White House in Washington, DC, US, on Thursday, April 10, 2025.

Photo:

Data Centers Now Need A Reactor’s Worth Of Power, Dominion Says

Robot cop at New Year celebration in Thailand.

The MareNostrum 5 supercomputer in Barcelona.

An image of an AI generated walking TV holding a flower.

Jblflip6

Eufysolocam

Alicia Witt in Urban Legend

Hp14

U.S. President Donald Trump speaks to the media during a guided tour of the John F. Kennedy Center for the Performing Arts before leading a board meeting on March 17, 2025 in Washington, DC.