Michelangelo

Title:
I built a tool to audit AI output — curious if this is a real problem

Post:
AI doesn’t fail loudly anymore. It fails subtly.

I kept seeing outputs that looked completely fine — but when I checked them, there were small things off. Numbers without sources, claims that couldn’t be verified, things that felt right but weren’t grounded.

So I built a simple tool:

• paste AI output

• it breaks it into claims

• flags what’s worth checking

• gives a trust score

Not trying to replace anything — just a quick sanity check before you rely on something.

You can try it here:
pinkhouse.tech/truth.html

If you want to test it freely:
Code: IH100

Curious if this resonates with anyone else here, or if I’m solving something niche.