@AnatoliKopadze - Karpathy Shows Five Daily

Open original

@AnatoliKopadze on X Andrej Karpathy walks through how he uses AI day to day, covering which model to pick, when thinking models pay off, turning one prompt into a full research report, running code through the model, and getting it to remember you across chats. The post is a pitch for a Claude-focused guide that expands on those same workflows.

  • Karpathy's two-hour talk breaks down model choice, when reasoning models are worth the wait, and how to get a full research report from a single prompt.
  • Two of the five segments cover letting the model run code for you and giving it persistent memory across separate chats.
  • The author claims most people use about 10% of these models' capability and frames the linked guide as covering the other 90%, mapping the same ideas onto Claude features.

The post is a promotional thread pointing to a video of Andrej Karpathy walking through his daily AI use, plus a follow-up guide the poster wrote about Claude features. The timestamps name five topics: which model to pick (18:03), when thinking models justify their extra cost (22:54), turning a single prompt into a full research report (42:04), having the model write and run code (59:00), and getting a model to remember you across separate chats (1:53:29). The author's pitch is that most people use about 10% of what these tools can do and the rest is what Karpathy demonstrates. The thread itself contains no detail beyond those timestamp labels. You won't find the actual model names, the cost tradeoffs, the prompt structure for research reports, or how the memory features work. Those specifics live in the video and in the linked guide, which the post promotes but doesn't summarize. So if you're deciding whether it's worth saving, the useful question is whether you want to sit through two hours of video or read the guide, since the thread gives you the agenda and not the answers. Treat the "90%" framing with some skepticism. It's a catchy hook, not a measured figure, and it's the kind of claim that tends to inflate the gap between casual and power use. The topics themselves are sound and widely useful, especially the question of when a reasoning model's slower, more expensive responses pay off versus when a fast model is enough. If you save this, save it for the video, and check the guide against Anthropic's own documentation before relying on any feature it describes, since third-party roundups of this kind can go stale or overstate what a feature does.

Ask about this article