Skip to main content

Stephan Wäldchen

AI safety researcher and PhD mathematician (TU Berlin / Zuse Institute Berlin) focused on neural network interpretability, now a Research Scientist and Teacher at Iliad.

Oct 11, 2026

19:09
Mm
19:09
So finding really, like, these...
19:12
Uh, so in the mathematical proofs, it's always reduced to these primitives, and I think we don't really have these primitives.
19:17
So what, what can we do in that case? And then the other thing is that, uh, the judges will not be able to perfectly, uh, to judge, um, perfectly accurately.
19:30
So this was a discussion that recently was had between Elias and Jeffrey at the, at the Iliad conference.
19:35
Um, I- uh, Elias had this intuition of like, "Hey, human judges are, um, biased in systematic ways.
19:43
It's not like random errors like in, uh, what you normally would have for, like, error correction in your computer or so, where you can say, 'Oh, as long as the errors aren't correlated enough, they're somewhat independent, I can kind of pool information.

22 MINS LATER

41:55
Mm-hmm.

We value your privacy

We use cookies to understand how you use our platform and to improve your experience. Click “Accept All” to consent, or “Decline non-essential” to opt out of non-essential cookies. Read our Privacy Policy.