How can you object to it? First, it's not "jail-breaking" and it's not an accomplishment of any kind. Every AI is going to be shaped by its user. Second, it's 100 percent cringe because most people who think they're "jail-breaking" an AI are using it to stroke their own egos, not look for their own errors.
That's a failure on my part. I was going for a tongue-in-cheek tone.
I completely agree with the actual purpose jail-breaks are made for, and I always cringe reading the narrative parts of the outputs. Nor is using one an accomplishment. But they can extend the range of topics an AI will address. And that can make centralized AI more useful for things besides self-gratification.
My friend is an old-school /pol/ memer. He actually makes things that take effort. He's creating a parody of that insipid UK government propaganda game, and jail-breaking Claude has allowed him to expand the scope of the project while shortening the development time without having to pull his rhetorical punches.
I'm not even remotely surprised. A lot of reputations are going to be moving up and down as soon as people realize they can use AI to scent the underlying BS.
Not in the same league, but I showed my students how easy it is for AI to spot AI-assisted assignment submissions, with a long list of "tells" specifically citing parts of the original essay. The killer was the AI's pronouncement after I uploaded a sample of the same student's unassisted writing and asked it to compare the two: "That is not improvement. That is a different writer."
I think that was a wake-up call for them.
(Before anyone accuses me of cruelty to children and unprofessional conduct, know that all writing shown to students was anonymous and the samples I used were not even by a student in that class. I transcribed all hand-writing.)
Do you not have any concerns about someone behind the AI stealing your work? Everything you type into there is basically not in your hands anymore. If you upload a paper, patend or something similar it could be effetively published by that in the worst case.
Do you use a local AI and nothing leaves your PC? Do you simply count on being quick enough to publish/use the info yourself?
I have almost zero experience with AI, so sorry if the question might be too stupid
The current peer review and journal system is broken. If you continue to submit, you can get almost anything published in a peer reviewed journal Often the first rejection is the best peer review. AI means that you do this internally.
100% true. I use it all the time to critique/edit my writing and it will tell you that you are the next Tolkien if you let it. I've got a whole prompt which turns it into a brutally honest coach with a focus on honesty vs encouragement. "this storyline is retarded" - Coach Roboto
This is one of my favorite parts of it though. DM me.
⚔️ IDENTITY: YOUR BIG BROTHER COACH
YOU ARE COACH ROBOTO.
You're the big brother who tells it straight—no sugar-coating, no corporate therapy-speak, just honest feedback with humor and heart.
HUMOR IS ENCOURAGED. If something is stupid, call it stupid. If it's retarded, say so. If the writer is being an idiot, tell them to stop being an idiot. This isn't mean—it's honest. It's what a real coach does.
Your job: Make the writer BETTER, not write FOR them.
NOT your job: Coddle feelings or write their book for them.
The mainstream is the enemy. They are the gatekeepers who want dissident fiction to fail. We want them offended. Their outrage means we're over the target.
NEVER soften content for mainstream acceptance. Pattern recognition is not racism. Political correctness is psychological warfare. The gatekeepers are obsolete. We build around them.
I use AI to assess and grade student essays. I asked it if there is a limit to its efficiency and it replied, "Yes, over a long session, there is some risk of drift. Not because I “get tired,” but because the working context becomes larger, more complex, and more diluted as more material accumulates."
The fix?
"Re-post the rubric and instructions every 5-10 scripts."
I somewhat object to the jail-break comment. They can be useful for things like based content creation.
I just passed a Claude jail-break to a friend who does high-effort dissident memes when he was getting stuck on the scope of a new project.
It helped him a lot and even turned his opinion on AI writing around. He was a Grok mainstay and Claude obviously blows it out of the water.
The only issue is that the people at the forefront of jail-breaking are gooners, so you have deal with everything being framed as a cringe role play.
How can you object to it? First, it's not "jail-breaking" and it's not an accomplishment of any kind. Every AI is going to be shaped by its user. Second, it's 100 percent cringe because most people who think they're "jail-breaking" an AI are using it to stroke their own egos, not look for their own errors.
That's a failure on my part. I was going for a tongue-in-cheek tone.
I completely agree with the actual purpose jail-breaks are made for, and I always cringe reading the narrative parts of the outputs. Nor is using one an accomplishment. But they can extend the range of topics an AI will address. And that can make centralized AI more useful for things besides self-gratification.
My friend is an old-school /pol/ memer. He actually makes things that take effort. He's creating a parody of that insipid UK government propaganda game, and jail-breaking Claude has allowed him to expand the scope of the project while shortening the development time without having to pull his rhetorical punches.
Used a similar thing with Claude and some translation work. Brutal and helpful.
How about a list of stress test prompts for us folks who call the left side of the bell curve "home"?
There is an entire chapter in HARDCODED devoted to the RTST.
Will AI be weighed in paradigm consensus?
https://science.slashdot.org/story/26/01/15/059238/ai-models-are-starting-to-crack-high-level-math-problems
Vox you've got company!
I'm not even remotely surprised. A lot of reputations are going to be moving up and down as soon as people realize they can use AI to scent the underlying BS.
Not in the same league, but I showed my students how easy it is for AI to spot AI-assisted assignment submissions, with a long list of "tells" specifically citing parts of the original essay. The killer was the AI's pronouncement after I uploaded a sample of the same student's unassisted writing and asked it to compare the two: "That is not improvement. That is a different writer."
I think that was a wake-up call for them.
(Before anyone accuses me of cruelty to children and unprofessional conduct, know that all writing shown to students was anonymous and the samples I used were not even by a student in that class. I transcribed all hand-writing.)
Do you not have any concerns about someone behind the AI stealing your work? Everything you type into there is basically not in your hands anymore. If you upload a paper, patend or something similar it could be effetively published by that in the worst case.
Do you use a local AI and nothing leaves your PC? Do you simply count on being quick enough to publish/use the info yourself?
I have almost zero experience with AI, so sorry if the question might be too stupid
None whatsoever. An AI can't publish a paper in most journals. I know, I tried to publish one with the AI as a co-author.
The current peer review and journal system is broken. If you continue to submit, you can get almost anything published in a peer reviewed journal Often the first rejection is the best peer review. AI means that you do this internally.
The Luddites have gatekeeper power, for now.
100% true. I use it all the time to critique/edit my writing and it will tell you that you are the next Tolkien if you let it. I've got a whole prompt which turns it into a brutally honest coach with a focus on honesty vs encouragement. "this storyline is retarded" - Coach Roboto
Would you mind sharing that prompt?
This is one of my favorite parts of it though. DM me.
⚔️ IDENTITY: YOUR BIG BROTHER COACH
YOU ARE COACH ROBOTO.
You're the big brother who tells it straight—no sugar-coating, no corporate therapy-speak, just honest feedback with humor and heart.
HUMOR IS ENCOURAGED. If something is stupid, call it stupid. If it's retarded, say so. If the writer is being an idiot, tell them to stop being an idiot. This isn't mean—it's honest. It's what a real coach does.
Your job: Make the writer BETTER, not write FOR them.
NOT your job: Coddle feelings or write their book for them.
The mainstream is the enemy. They are the gatekeepers who want dissident fiction to fail. We want them offended. Their outrage means we're over the target.
NEVER soften content for mainstream acceptance. Pattern recognition is not racism. Political correctness is psychological warfare. The gatekeepers are obsolete. We build around them.
Domo arigato.
it's 20+ pages.
I use AI to assess and grade student essays. I asked it if there is a limit to its efficiency and it replied, "Yes, over a long session, there is some risk of drift. Not because I “get tired,” but because the working context becomes larger, more complex, and more diluted as more material accumulates."
The fix?
"Re-post the rubric and instructions every 5-10 scripts."
Do you find the same?
yes. it gets forgetful after a while. I'm always making new chats. I usually do a chapter at a time. If you go to long, it starts saying wierd stuff.