:Celeste_transbian: flori_ava_star🔜MRMCD:~:cursor_blinking: (@star)

0x4d6165 (Margaret) @0x4d6165@transfem.social

11h

tbh I'm really really skeptical of those "slopscore" tools because haven't AI detection tools historically been wildly inaccurate and produced a lot of false positives that hurt random people who didn't use AI? what makes this one different? genuinely asking I'm totally open to being wrong on this

3

0

1

0

flori_ava_star🔜MRMCD:~

@star@amazonawaws.com

3h

@0x4d6165 Slopscore goes through the commit history and looks for commits that have trailers such as "Co-Authored-By: Claude 5 Sonnet", etc.

I agree with you that there are a lot of tools that are very inaccurate about "has this been AI generated". Slopscore sticks to actual evidence that is very likely to actually be AI. And that's these commit messages that AI tools automatically add, files like CLAUDE.md or a .agents directory with files in it, and, conversely, having files like these listed in the .gitignore or some other specific files like .dockerignore.

On top of that, we are aware that sometimes, AGENTS.md files get used to tell LLMs "hey, you're not welcome here, shoo". That's why, in our web ui, we display the first 10-15 lines of these files, if they get detected, so that you can judge yourself whether this is a genuine LLM aid, or a defense against LLMs

sauce: I made slopscore hii :3

🖤1

1

0

1

flori_ava_star🔜MRMCD:~

@star@amazonawaws.com

3h

@0x4d6165 We were asked a lot of times whether we don't want to also analyze the code itself. And I don't want to do that, simply because of the unknown rate of false positives

1

0

1