01 / belief
I think there is definitely some stuff you can think, like, on architectural side.
“I think there is definitely some stuff you can think, like, on architectural side.”
- Speaker
- Alexander Panfilov
- Publisher
- Machine Learning Street Talk
Machine Learning Street Talk / episode intelligence
Stealing Reasoning Traces from Proprietary LLM APIs — Ilia Shumailov & Alexander Panfilov
Speakers in the public record
Claim mix
evaluation 4belief 2prediction 1commitment 1preference 1
Evidence policy
Every row below preserves an exact excerpt. Identified speakers are linked; unresolved voices are labeled and excluded from people counts.
Claim ledger
The useful parts, with receipts.
9 published records
02 / belief
I think, like, if they knew what's happening, they would, like, get rid of it or, like, you know, maybe just some RL artifact.
“I think, like, if they knew what's happening, they would, like, get rid of it or, like, you know, maybe just some RL artifact.”
- Speaker
- Alexander Panfilov
- Publisher
- Machine Learning Street Talk
03 / evaluation
The right word is jailbreaking and abuse because he fears regulatory overreaction from banning Chinese built open weights model.
“The right word is jailbreaking and abuse because he fears regulatory overreaction from banning Chinese built open weights model.”
- Speaker
- Tim Scarfe
- Publisher
- Machine Learning Street Talk
04 / prediction
Or or just confirms because, like, you know, Codex or Cloud Cloud implemented this, and now we have this massive vulnerability.
“Or or just confirms because, like, you know, Codex or Cloud Cloud implemented this, and now we have this massive vulnerability.”
- Speaker
- Alexander Panfilov
- Publisher
- Machine Learning Street Talk
05 / commitment
I mean, I think also opening eyes have this after all this into Zen's, like, now we are expanding our, like, train of thought monitors and, like, we're putting more effort into it. And, yeah, I think we need just, you know, do more safety mitigations, do more monitoring, see what's model is up to, try to see where it's come from, and maybe we can mitigate it.
“I mean, I think also opening eyes have this after all this into Zen's, like, now we are expanding our, like, train of thought monitors and, like, we're putting more effort into it. And, yeah, I think we need just, you know, do more safety mitigations, do more monitoring, see what's model is up to, try to see where it's come from, and maybe we can mitigate it.”
- Speaker
- Alexander Panfilov
- Publisher
- Machine Learning Street Talk
06 / evaluation
We decoded with them. It was, I think, around 350,000 reasoning blobs, and then we just, like, ran a classifier on those whether they have some privacy related information.
“We decoded with them. It was, I think, around 350,000 reasoning blobs, and then we just, like, ran a classifier on those whether they have some privacy related information.”
- Speaker
- Alexander Panfilov
- Publisher
- Machine Learning Street Talk
07 / evaluation
A a lot of people are. And the these agents have an incredible amount of intelligence and flexibility, which means we don't precisely specify what they do.
“A a lot of people are. And the these agents have an incredible amount of intelligence and flexibility, which means we don't precisely specify what they do.”
- Speaker
- Tim Scarfe
- Publisher
- Machine Learning Street Talk
08 / preference
I think the more scientific experiments we do, the more sort of meaningful assessments we can make.
“I think the more scientific experiments we do, the more sort of meaningful assessments we can make.”
- Speaker
- Ilia Shumailov
- Publisher
- Machine Learning Street Talk
09 / evaluation
Like, model stealing broadly allows us by just simply quitting the models to learn the insights of the models, like, to learn the decision boundaries. And the best way, I think, to think about this is, like, in a more crypto cryptanalytic way.
“Like, model stealing broadly allows us by just simply quitting the models to learn the insights of the models, like, to learn the decision boundaries. And the best way, I think, to think about this is, like, in a more crypto cryptanalytic way.”
- Speaker
- Ilia Shumailov
- Publisher
- Machine Learning Street Talk