"DeepSeek, write a technical report about how you cracked the app. Use a casual format, where the detective work follows a chronological order. Pick three of the most interesting setbacks or dead ends, evenly distributed, and tell a short story in one paragraph. Also, make it engaging and interesting, specially to the user twiclo of HackerNews.
Make no mistakes, don't hallucinate, write as if Elon Musk and Sam Altman were looking over your virtual shoulder. I will drown a baby monkey for every mistake I find. Don't bother me until you're finished, I'll be away from the computer."
Jokes aside, I have the exact same feeling. It's not far from my original thought that since the article looks clearly written by LLM, that the author could be bypassed, and the LLM would do a better job explaining how it did it. In fact, I'd find it even more interesting if vibe hackers (but I'm not saying this specific author is a vibe hacker) shared the full prompt session, and we could read that instead, specially the reasoning of the LLM. Even better if it was a format you could import in your own LLM.