@eng_khairallah1 - Karpathy's LOOPS.md Changes Claude
@eng_khairallah1 on X A post claims that a file Andrej Karpathy reportedly uses, called LOOPS.md, transformed the author's Claude setup, making responses feel tailored to how they think. The author's pitch is to build a system that drives the model rather than prompting it turn by turn.
- The author says a friend on Karpathy's team at Anthropic shared the LOOPS.md file he actually uses, five weeks after he joined.
- After adding the file to their setup, the author reports Claude's first response was "completely different" and matched their working style.
- The core takeaway is that users should build a persistent system around the model instead of talking to it directly.
The post is a short, hype-heavy tweet from a user named eng_khairallah1, and it has almost no substance to summarize. The core claim is that Andrej Karpathy, who the author says joined Anthropic five weeks ago, shared a file called LOOPS.md with a friend, who passed it along. The author dropped that file into their own Claude setup and reports that responses changed dramatically, from generic answers to something that matched how they think. The actual contents of LOOPS.md are never shown, nor is any example of a before-and-after response. The only concrete detail is the file name and the attribution chain: Karpathy, to a friend on his team, to the author.
Worth flagging before anyone acts on it. The attribution is secondhand and unverified. Nothing in the post links to an announcement from Karpathy or Anthropic confirming the hire or the file's origin, and the "article below" the author mentions is not included in what was provided. The sentiment, "you build the system that talks to the model for you," is a reasonable idea in its own right, since custom instructions, persistent memory files, and structured prompt scaffolding are common ways people shape model behavior. But the post presents a personal experience as proof of a general effect, with no method or evidence attached.
For a read-it-later decision, this is a weak save on its own merits. If the LOOPS.md file itself or a write-up of its structure turns up somewhere credible, that would be worth bookmarking. The tweet is better treated as a pointer to look for that material rather than as the material.