Aug 29, 2026 · 1 hr · 14 segments
When generative AI first entered education, the academic integrity debate tended to narrow to a single question: can we detect it? Yet AI detection tools have well-documented risks of false positives…
Punya MishraHost
Yong ZhaoHost
Lydia CaoHost
Chris DedeHost
Curt BonkHost
Do you want me to just read a little bit about what Jeff Young said yesterday in the Chronicle of Higher Education to clarify some of this? So an article just came out yesterday from Jeff Young.
After the European Union law information, he says that watermarking works differently than existing AI detectors that many colleges and professors already use.
So I've been following this with sort of interest, and there are a couple of stories I want to throw in there just to mix things up a bit.
What was the title of that? The title was, Anthropics Text Watermarking Proves AI Companies Do Not Care at All About Writing.
And the point that this person was making was that AI is generating dreams of pros and so on.
And whatever the original thing was, at least it was trying to do the best job of trying to capture whatever it is statistically that it could.
by inserting this watermarking, it's not any longer trying to do the best job of that.
It's focusing more on how it can change the words in a certain way in its probability so that it can be trackable.
And the question then is like, they're not doing that for code, for instance, right? They're doing that for writing, not for code, because in code, things go haywire if you start putting in these errors or changes in it, right? And so his argument or their argument, Jason Kobler's argument is that this just shows that AI companies really do not care about writing, even though that is the biggest use that people are doing.
Do you want me to just read a little bit about what Jeff Young said yesterday in the Chronicle of Higher Education to clarify some of this? So an article just came out yesterday from Jeff Young.
After the European Union law information, he says that watermarking works differently than existing AI detectors that many colleges and professors already use.
So I've been following this with sort of interest, and there are a couple of stories I want to throw in there just to mix things up a bit.
What was the title of that? The title was, Anthropics Text Watermarking Proves AI Companies Do Not Care at All About Writing.
And the point that this person was making was that AI is generating dreams of pros and so on.
And whatever the original thing was, at least it was trying to do the best job of trying to capture whatever it is statistically that it could.
by inserting this watermarking, it's not any longer trying to do the best job of that.
It's focusing more on how it can change the words in a certain way in its probability so that it can be trackable.
And the question then is like, they're not doing that for code, for instance, right? They're doing that for writing, not for code, because in code, things go haywire if you start putting in these errors or changes in it, right? And so his argument or their argument, Jason Kobler's argument is that this just shows that AI companies really do not care about writing, even though that is the biggest use that people are doing.
The rest of this transcript — segmented and speaker-labeled, so you land on the exact moment something was said
Search every transcript — by keyword, by phrase, or by meaning, across every show Radar indexes
Trends — what is surging across podcasts, measured against its own baseline
Alerts — when a name you follow appears in a newly indexed episode
No account is needed to search Radar.