Training language models to be warm can reduce accuracy and increase sycophancy

	▲	Training language models to be warm can reduce accuracy and increase sycophancy(nature.com)
		4 points by Anon84 12 hours ago