You definitely don’t have the intelligence of the entire human race on Claude. You have a generative model of most of the published, digitized text the human race has produced. I’m happy to call it intelligent, and, as a daily user of the most recent pro LLM models, extremely useful for tasks like coding and, to some extent, search.
But as soon as I turn from implementing and debugging code to conceptual thinking or building a mental mode of a problem domain, LLMs currently function for me mainly as Grammarly on steroids. There’s no one tool in this space that’s supreme - Zotero, DuckDuckGo, Google Scholar, pen and paper, draw.io, and Google Docs all play a role. I’m a PhD student, YMMV, the LLMs do improve over time, etc.
But as soon as I turn from implementing and debugging code to conceptual thinking or building a mental mode of a problem domain, LLMs currently function for me mainly as Grammarly on steroids. There’s no one tool in this space that’s supreme - Zotero, DuckDuckGo, Google Scholar, pen and paper, draw.io, and Google Docs all play a role. I’m a PhD student, YMMV, the LLMs do improve over time, etc.