@bilalchughtai_ profile picture
bilal ๐Ÿ”ถ @bilalchughtai_

agi safety research @ google deepmind | cambridge mmath

Joined May 2015
  • Posts 261
  • Following 693
  • Followers 1,186
27 Photos and videos
Pinned Post
bilal ๐Ÿ”ถ@bilalchughtai_
Jun 12
New research update from the Google DeepMind Language Model Interpretability team. We build and evaluate dead simple open-ended model diffing agents tasked with studying the behavioural differences between two models, and find them to be promising in practice.
4
12
135
34,463
RT @MaryPhuong10: We're releasing the GDM AI Control Roadmap -- our plan for building internal security against potentially adversarial AIโ€ฆ
42
6
bilal ๐Ÿ”ถ reposted
Josh Engels@JoshAEngels
Jun 19
Text diffusion models are fast, but are less transparent than today's LLMs because they do many forward passes before outputting text. We audit the transparency of DiffusionGemma and find that the intermediates are interpretable. This recovers many of the benefits of CoT! ๐Ÿงต
3
47
246
65,928
bilal ๐Ÿ”ถ reposted
Josh Engels@JoshAEngels
Jun 13
New GDM interp research: SFT is a big deal for safety relevant behaviors. We recently investigated root causes for some of Geminiโ€™s behaviors. We were surprised to find that many behaviors actually came from the initial supervised finetuning stage, not later stages like RL! ๐Ÿงต
5
34
267
99,266
New research update from the Google DeepMind Language Model Interpretability team. We build and evaluate dead simple open-ended model diffing agents tasked with studying the behavioural differences between two models, and find them to be promising in practice.
4
12
135
34,463
We think more mature versions of tools of this form may play a useful role in understanding and improving model behaviour. For instance, they may help answer the question: what is the behavioural effect of training models with subtly different constitutions?
14
1,101
More details in the blogpost: lesswrong.com/posts/qi4mNbZYโ€ฆ Work with @JoshAEngels and @NeelNanda5

Building and evaluating model diffing agents โ€” LessWrong

This is the second in a series of informal research updates from the Google DeepMind Language Model Interpretability team, in interpretability and adโ€ฆ

lesswrong.com
24
1,052
bilal ๐Ÿ”ถ reposted
Reilly H@ReillyHaskins02
May 27
Could future models learn that their CoT is being monitored and hide their reasoning to evade detection? In our new paper, @JoshAEngels, @bilalchughtai_, and I find that yes, models finetuned on docs describing a CoT monitor evade detection far more often than unaware models ๐Ÿงต
2
11
66
8,654
why i changed my mind on what to do with completed tasks: bilalchughtai.substack.com/pโ€ฆ
7
1,104
why i run some lessons from the past year on the road bilalchughtai.substack.com/pโ€ฆ
6
363
overnight oats are awesome bilalchughtai.substack.com/pโ€ฆ
4
353
bilal ๐Ÿ”ถ reposted
Arthur Conmy
@ArthurConmy
Jan 19
Our new @GoogleDeepMind paper studies novel activation probe architectures for classifying real-world misuse risks. Our research has informed live deployments of probes in Gemini. ๐Ÿงต
16
59
742
157,561
bilal ๐Ÿ”ถ reposted
Jordan Taylor@JordanTensor
9 Dec 2025
NEW PAPER from UK AISI Model Transparency team: Could we catch AI models that hide their capabilities? We ran an auditing game to find out. The red team built sandbagging models. The blue team tried to catch them. The red team won. Why? ๐Ÿงต1/17
2
22
111
26,098
bilal ๐Ÿ”ถ reposted
Samuel Marks@saprmarks
6 Dec 2025
Research related to detecting AI deception has a bunch of footguns. I strongly recommend that researchers interested in this topic read GDM's position piece documenting these footguns and discussing potential workarounds. More reactions in ๐Ÿงต
bilal ๐Ÿ”ถ@bilalchughtai_
3 Dec 2025
can we detect when an AI system is being strategically deceptive without relying on behavioural evidence? we, on the GDM mechanistic interpretability team, spent many months working on this problem in our new position piece, we share what we learned:
3
5
38
3,387
can we detect when an AI system is being strategically deceptive without relying on behavioural evidence? we, on the GDM mechanistic interpretability team, spent many months working on this problem in our new position piece, we share what we learned:
5
14
150
33,743
finally, we discuss a further problem that might pose a roadblock to building deception detectors even if the problem of label assignment is solved; namely, that there may not exist a universal set of mechanisms that produce or pinpoint deception in LLMs.
1
22
978
arxiv: arxiv.org/abs/2511.22662 authors: lewis smith, myself, @NeelNanda5

Difficulties with Evaluating a Deception Detector for AIs

Building reliable deception detectors for AI systems -- methods that could predict when an AI system is being strategically deceptive without necessarily requiring behavioural evidence -- would be...

arxiv.org
1
1
29
965
Load more