Spotting AI-written text

Truesight

Read the text. Don't trust a score blindly.

Four ways of looking at the same words.

Most detectors give you one number from one hidden model. Truesight runs four that work in completely different ways, and shows you what each one sees, sentence by sentence.

Desklib
A model trained to tell human and AI writing apart.
The committee met on Tuesday to look over the budget. Several members felt the proposal was thin on detail. In conclusion, it is important to note that careful planning ensures sustainable outcomes for all stakeholders.
GLTR
How predictable each word was. Green is very predictable, red is surprising.
The committee met on Tuesday to look over the budget. Several members felt the proposal was thin on detail. In conclusion, it is important to note that careful planning ensures sustainable outcomes for all stakeholders.
Fast-DetectGPT
How closely the text matches what an AI would most likely have written.
The committee met on Tuesday to look over the budget. Several members felt the proposal was thin on detail. In conclusion, it is important to note that careful planning ensures sustainable outcomes for all stakeholders.
ImBD
Signs that human writing has been polished by an AI.
The committee met on Tuesday to look over the budget. Several members felt the proposal was thin on detail. In conclusion, it is important to note that careful planning ensures sustainable outcomes for all stakeholders.

When they agree, it means something.

One flag on its own is only a reason to look closer. Several methods pointing at the same sentence is worth paying attention to.

78%
Desklib
41%
GLTR
66%
Fast-DetectGPT
83%
ImBD

Your text never leaves your machine.

It runs entirely on your own computer. Free and open source, with nothing held back behind a paid tier.

See on GitHub
Truesight is made by SyncraNext: Finalist →