Will OremusJan 31, 2023

OpenAI's new tool to detect AI-written text correctly spotted 26% of AI-written text, while falsely identifying 9% of human-written text as AI-written.

One way of interpreting this: AI isn't yet good enough to reliably identify AI content.

Another: The technology to detect AI text is lagging well behind the technology to produce it. And/or it turns out to be a harder problem. https://openai.com/blog/new-ai-classifier-for-indicating-ai-written-text/

New AI classifier for indicating AI-written text

We’re launching a classifier trained to distinguish between AI-written and human-written text.

Show threadMarcel Salathé

@willoremus It is practically useless, for reasons outlined here: https://digitalepi.substack.com/p/how-useful-is-openais-new-ai-detection

See also https://mastodon.social/@marcelsalathe/109787200129188289

How useful is OpenAI's new AI detection tool?

As we've learned during the early phase of the COVID pandemic, detecting rare things requires accurate tests.

Digital Epidemiology
Feb 1, 2023 at 2:50amElk