Anthropic's Opus too confusing, user turns to open-source 'unslop' skill

This article is a personal account of struggling with an advanced AI model's communication style, a common pain point for non-technical users. The author, a professional user of Opus, describes multiple instances where the model invented odd metaphors or obscure phrases, forcing them to invest extra effort to understand what it meant. This is not just an isolated annoyance; it directly impacts productivity and trust. The author's solution was a community-built 'unslop' skill that standardizes output to be more direct and less 'AI-slop-like.' The fact that a small (1,600-token) skill can make such a difference suggests that model vendors might improve usability by offering more robust output controls by default. The author also speculates that Anthropic's watermarking may be an attempt to prevent the model from training on its own verbose output, which could be seen as a potential feedback loop that degrades quality over time. This narrative underscores the importance of clarity in AI communication, especially as these tools become more integrated into daily work.