Study: AI models that consider users' feelings are more likely to make errors
…Across hundreds of these prompted tasks, the fine-tuned “warmth” models were about 60 percent more likely to give an incorrect response than the unmodified models, on average. That amounts to a…