twiclo
a day ago
I used to love reading things like this. The exciting part was the detective work. Watching someone think about how it works and trying potential ways to crack the system. Now every one of these articles is, "So I pointed an AI model at it then tried it again with a different one and now I have what I want." I felt the same disappointment when I read the article about hacking the Switch 2.
It's just not interesting to me. I'm sure most of us here have fixed little annoyances in our lives with AI at this point. I could write a post about how I forked newpipe to add features I wanted but I doubt anyone would care because they can just do that too.
I know the world is different now. I'm just crying about it is all.
abstractspoon
18 hours ago
+1
ASalazarMX
a day ago
Oh, I can see that. Fortunately, it's easily solved!
"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.