Show HN: I'm building an AI-proof writing tool. How would you defeat it?

Posted by callmeed 5 hours ago

Counter7Comment8OpenOriginal

Comments

Comment by ephou7 1 hour ago

I used this:

#!/usr/bin/python3 import subprocess import time import random with open("/tmp/x") as f: t = f.read() for c in t: subprocess.call([ "xdotool", "type", c ]) time.sleep(abs(random.gauss(0,0.07)))

And pasted a random Hacker News comment:

Authenticity Score 81 Highly Authentic

Words per minute: 162 Keystroke variance: 52ms Paste attempts: 0 Window/tab switches: 4 Pauses (≥10s): 0 DOM manipulations: 0

You failed.

Comment by ticulatedspline 4 hours ago

- Screencap and OCR the prompt

- feed it to an llm to answer

- use a tool [1] to emulate human typing cadence.

- Use a tool to send mouse events to the browser window.

[1] https://multilogin.com/blog/paste-as-human-typing/

not sure if you have any feedback at the end but if you do the alternative is feedback poisoning in your training set to mark AI stuff as real and real stuff as fake. The former is automatable and if there were a good reason you could probably mechanical-Turk a large number of people at low cost to do it for real and give feedback that they were actually AIs.

Comment by nine_k 1 hour ago

An AI-proof writing tool is a pen and paper, on a glass table, in a Faraday-cage locked room, etc, etc.

Depending on how high are the stakes, weeding off just the most obvious cases of AI usage may be enough. But wherever an electronic input is involved, it can be emulated in ways that are impossible to detect on a reasonable budget. Content analysis may help somehow, but again it would only detect the more obvious cases.

Comment by forgotpwd16 2 hours ago

Or write a script that does all that and bind it to shortcut. Eventually https://files.catbox.moe/zargud.png. (The window/tab switches were 1~2 when ended; reached 9 after few attempts to screenshot it.)

Comment by dpoloncsak 2 hours ago

>Pasting and DOM manipulation are disabled to ensure all writing is original.

>We track telemetry such as typing speed, pauses, tab changes, and window focus events.

People figure out ways around this for like...Runescape bots and other low-stake situations. I don't think it would hold up to anything other than casual users. Seems like an agent could whip something up in Auto-HotKey or something.

I get this is the extreme end, but if this gets popular enough, can't you write like a custom 'keyboard' driver that just takes AI input and 'types' it? Random delay between keystrokes, whatever....

It also can't be used to verify existing work, right? I can't see if a student's essay is LLM-written. Is there any real-world use you see? Or is this just a fun toy?

Comment by yjftsjthsd-h 1 hour ago

> I get this is the extreme end, but if this gets popular enough, can't you write like a custom 'keyboard' driver that just takes AI input and 'types' it? Random delay between keystrokes, whatever....

We can easily go one more step than drivers; making a cheap microcontroller enumerate as a USB keyboard is easy.

Comment by dpoloncsak 1 hour ago

Ai powered rubber duckies go crazyyyyyyy

Comment by stupidgeek314 1 hour ago

> Open chatgpt in second tab > Type what chatgpt says yourself

The only "AI proof" writing tool is those blue books you take exams on in college.