This article is just reporting a few results from Anthropic's blog. The original blog post is much more interesting and contains nice visualizations: https://www.anthropic.com/research/claude-values-models-lang...

Claude kind of sucks, to be honest.