/Tech1h ago

Alex Imas, Google DeepMind Director of AGI Economics, finds AI detectors maintain low false positive rates, drawing skepticism from Miles Brundage

Brundage argues error rates still exceed one in 10,000

3700220

Original post

@alexolegimas @honorablepicnic Hmm, much higher than 1 in 10k... in any case, my claim has never been that it's not useful, but I think many people have certainly gotten ahead of themselves. E.g. claiming with not a hint of uncertainty that the Pope is putting out fully AI generated tweets

Alex Imas@alexolegimas

@Miles_Brundage @honorablepicnic We test both long passages and short passages. False negatives do increase with shorter text, but false positives are pretty stable and low

4:26 PM · Jun 25, 2026 · 73 Views

Sentiment

Negative users called claims about Pangram maintaining the lowest false positive rates on short AI text pretty wild and expressed skepticism toward the reported performance.

Pos

0.0%

Neg

100.0%

1 comments with sentiment.

Cluster Engagement

Digg Deeper

No Digg Deeper questions have been answered for this story yet.

Posts from X

Most Activity

VIEWS20LIKES1

Alex Imas@alexolegimas

@Miles_Brundage @honorablepicnic I agree with that. There is a strong use case for repeat offenders but this is pretty wild.

1h201