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 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.
OpenAI has officially announced its next-generation foundation model GPT-5, achieving major breakthroughs in reasoning, mathematics, and code generation, with benchmark scores surpassing all previous models.