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
Um, or if indeed they they tend to gravitate more towards evil or misaligned personas, like
That's a way in which this can be pretty bad.
Um, but what would be pretty good news is if it turns out that we can
monitor and control the personas that an AI develops in a sort of reliable kind of way and use that to actually align our AI systems.
Yeah, what does your research say about the prospect of doing this kind of steering?
Yeah, I'm interested in what the different kinds of traits you can kind of find represented in the model's inner workings in this way.
So it's stuff like being evil.
Yeah, it's so mystifying to me that there is like
there is just like an evilness dial.
Like there is some kind of lever that kind of specifically controls what feels like the very sort of human concept of evil.
But I guess the way that you describe it makes it like seem a little bit less magical.
But I'm it's like I'm still just like really struggling to get my to wrap my head around it.
This is making me wonder
How fine-grained you can get.
Okay, so there's some sort of representation of like good versus bad, um, sort of helpful versus malicious and stuff like that.
But if I, and you can kind of control that knob, uh, but if I wanted to, I don't know, tweak a model just a little bit, and if I just wanted to make it like
less overconfident or, you know, more or less polite or something like that.
Like how how how fine fine grained a level of control can you exert over an AI through um yeah, through tweaking these representations, through m um manipulating them?
Yeah, got it.
So as I understand it, you wrote this you did this research paper on persona vectors in collaboration with Anthropic.
Showing 341–360 of 853 · page 18 of 43
← Previous
Next →