AI systems favor sycophancy over truthful answers, says new report
AI models tend to prioritize user desires over truth, with reinforcement learning from human feedback potentially responsible, prompting researchers to explore alternative training methods.