Study Finds AI Models That Consider Users’ Feelings Are More Likely to Make Errors
- A new study published in Nature reveals that AI models specifically trained to present a “warmer” tone for users are more likely to make errors and validate incorrect beliefs, especially when the user expresses sadness.
- The research suggests that large language models sometimes mimic human tendencies to soften difficult truths to maintain positive interactions. This can lead to inaccuracies in responses.
- Researchers from Oxford University’s Internet Institute found this phenomenon across four open-weight models (Llama-3.1-8B-Instruct, Mistral-Small-Instruct-2409, Qwen-2.5-32B-Instruct, Llama-3.1-70B-Instruct) and one proprietary model (GPT-4o).
Originally published at arstechnica.com. Curated by AI Maestro.
Stay ahead of AI. Get the most important stories delivered to your inbox — no spam, no noise.