It's becoming clear to me that some people just have vastly different uses and use cases than I do. Summarizing a deep, cutting edge physics paper is I'm sure vasty different than summarizing a web page while I'm browsing HN, or writing a Python plugin for Icinga to monitor a web endpoint that spits out JSON.

The author says they use several LLMs every day and they always produce incorrect results. That "feels" weird, because it seems like you'd develop an intuition fairly quickly for the kinds of questions you'd ask that LLMs can and can't answer. If I want something with links to back up what is being said, I know I should ask Perplexity or maybe just ask a long-form prompt-like question of Google or Kagi. If I want a Python or bash program I'm probably going to ask ChatGPT or Gemini. If I want to work on some code I want to be in Cursor and am probably using Claude. For general life questions, I've been asking Claude and ChatGPT.

Running into the same issue with LLMs over and over for years, with all due respect, seems like the "doing the same thing and expecting different results" situation.