
AI Aware
Detecting AI generated text, images, audio & deepfakes
As well as AI text detection we are testing deepfake video detection...
Combatting misinformation, fraud and AI-generated manipulation.
4 Likes
5 Comments
5 Comments
-
1
I tested your product and have a bit of feedback:
1. I took an AI-generated text and checked it through Deepfake Detection and GPTZero. Unfortunately, your product showed 0% AI, even though the text was fully written by AI. GPTZero flagged the same text as 100% AI.
Here's the text I used:
"Helen runs an app for expecting mothers. Last week she opened Achiv, and discovered subreddits where her audience was the most active, wrote and polished posts with Achiv extension. Five different communities. Five different writing cultures. Five different rule sets. Over 300 new users by the end of the week and growing.
The thing that makes this hard isn't one post. It's five.
Every subreddit is a different 'country'. One community tolerates promotion that another will downvote instantly. Some want long personal stories, others want a screenshot and a question. Some allow links, others remove anything with a URL in the first paragraph. Moderators on one sub will let you mention your product if it's relevant, moderators on another will nuke the post within thirty seconds for the same line. None of this is written in one place. It's accumulated culture, plus moderator preferences, plus AutoModerator rules you can't even see."
2. The font you used for "Choose text, image, video or audio", "Tips for best results", and "Use my feedback to help improve the AI model. Uncheck to opt out of training data collection." is very hard to read. I'd recommend replacing it.
-
1
Thank you for the feedback - it's very useful. What AI did you use to generate the text?
-
1
HiveMind (myosin.xyz/hivemind) — our AI strategy copilot. It's not publicly available yet, currently in closed beta, but I can drop you in with code HivemindIH123. That's probably worth it for your testing anyway — the output is generated through operator frameworks rather than generic prompts, which likely explains why the detection signal was weaker. The text uses contractions, rhythm shifts, and concrete numbers, and framework-based prompting tends to avoid parallel structures and tidy aphoristic closers that AI detectors typically flag.
If useful for calibrating your model, happy to run more samples through different prompt configurations. The harder detection problem is going to be exactly this kind of output going forward.
-
1
Could you email me at steve @ aiaware . io with either some more samples or way can generate this myself. Thank you
-
1
You can try it yourself at https://hivemind.myosin.xyz/auth/signup using the code HivemindIH123
-
-
-
-
About
The world is being flooded with AI-generated content.


Comment