This project is slop and doesn't work

[h3ndrik]

I feel the information on the README.md is a bit of a hit and miss.

• The Quick Start doesn't work, as there's no such image on the Docker Hub
• Discussions are linked but not enabled
• The "project board" leads to a dead link
• The Twitter link doesn't seem to be connected to the project?
• There's a lot of confusion on whether this repo is located at "vinpatel/Manuscript" or "manuscript/manuscript"
• ...

And then the actual returned JSON doesn't look like the one in the documentation? I don't get a "verdict", but a "human":true
And it's 37% confident a short ChatGPT snippet is written by a human.

Which makes me wonder about the accuracy of all the other info in the document. Is this project worth looking into? Are the benchmarks and other info accurate? Or made up by Claude Code?

[h3ndrik]

I managed to run the Python benchmark script. But I don't think any of this looks right. Contrary to the attempt in the repo, I let it analyze the AI samples first. And I get the following result on text:

Text Detection:
  Samples: 100 (TP=3, TN=10, FP=40, FN=47)
  Accuracy:  13.00%
  Precision: 6.98%
  Recall:    6.00%
  F1 Score:  6.45%

That is considerably worse than rolling a die?! !!!!!!!!!!!!!!!!!!