WEB Signal 581 3 feeds carried it
Rebuilt AI code comment classifier runs locally in the browser with 77 percent balanced accuracy
↗
Illustration only Photo by Declan Sun on Unsplash
A new AI code comment classifier has been rebuilt on public data and runs entirely in the browser, offering calibrated predictions and feature activation views.
Engineers can paste code comments into a web page to check if they were AI-generated without the text leaving their browser. The tool provides calibrated confidence scores, allowing users to gauge the reliability of each specific verdict rather than relying on aggregate accuracy alone.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
The classifier was rebuilt on public data, allowing the code and data to be shared unlike the previous private version.
It runs entirely in the browser, ensuring pasted code comments do not leave the user's machine.
The tool achieves 77 percent balanced accuracy on cross-validation and 88 percent accuracy on a smaller real-world test set.
THE CLUSTER