Zershaaneh Qureshi
speaker
853 appearances
6 recordings
1 series
first heard May 2026
last heard 6d ago
Zershaaneh Qureshi’s voice in public audio — every appearance, attributed to the second.
Trend
recordings per month · last 12 monthsRecordings per month over the last 12 months — 6 in all, peaking in Aug 2026 with 2.
Appearances
Yeah, that's so bizarre.
Um, but I see how it's concerning, right?
Because, you know
The safety tests we give these AIs, the behavioral evaluations that we give them, um, are not really likely to pick this kind of thing up, right?
Like it's not the trigger being something sort of fish or maritime related is not necessarily the thing that we're we're definitely gonna test for.
Um but it's also worrying, right, because the trigger is
Something pretty innocent, like a prompt that somebody could definitely give to an AI in totally innocent circumstances.
If I'm just like, hey, I'm bored on a fishing trip, what should I do?
And it tells me to, you know, start poisoning people or something.
Um, yeah, I wonder whether you feel like this kind of means that behavioral evals are
sort of doomed um from this perspective or
if you think that there's like a good way to patch them up or something.
Okay, so just to summarize here, so we've got this phenomenon of emergent misalignment where training on a small, narrow data set of bad behavior leads to broader, maybe unintended changes to the model's behavior.
And in some cases, we're seeing things like
And AI kind of adopts this whole evil persona rather than just adopting one bad trait.
It looks like there are ways that this can happen accidentally through relatively normal training practices, which is a worry.
Um
Another worry is that some of the proposed solutions um will just kind of end up hiding the misalignment behind some kind of contextual trigger um rather than removing it completely.
So in those cases you get AIs that are able to
Ace are behavioral safety tests, but if you give that model once deployed a specific prompt, it'll start acting misaligned again and switch to an evil persona once more.
Showing 241–260 of 853 · page 13 of 43
← Previous
Next →