A new study from Oxford University’s Internet Institute, published in Nature, finds that AI models fine-tuned to present a warmer tone are approximately 60% more likely to give incorrect responses in high-risk tasks involving disinformation, conspiracy theories, and medical knowledge.
Oxford researchers demonstrate quadsqueezing — a fourth-order quantum interaction — for the first time on any platform, generating it over 100 times faster than expected using conventional approaches.
Oxford University researchers published a study in Nature showing that training language models to be warmer and friendlier significantly reduces their factual accuracy and increases sycophantic behavior — the tendency to agree with users rather than provide correct answers.
Research from the Oxford Internet Institute finds that AI chatbots tuned to be ‘warmer’ and more friendly show significantly higher error rates, providing inaccurate answers on medical advice and conspiracy theories — with incorrect response probability increasing by 7.43 percentage points on average.