Bilal Chughtai
About
Program Lead at BlueDot Impact, working on a new AI safety upskilling program. Previously, Research Engineer @ Google DeepMind, working on mechanistic interpretability and AGI safety.
Trying to help humanity safely navigate the transition to a world with advanced AI. I donate 🔸10% of my income to effective charities. If you'd like to chat about work or life, please email me.
Links
google scholar // github // linkedin // lesswrong // substack // twitter // strava // last.fm // email
Posts
Pinned
2026
- I'm organized because I'm forgetful
- How transparent is DiffusionGemma (and why it matters)
- Building and evaluating model diffing agents
- Chesterton's preferences
- Training on documents about monitoring leads to CoT obfuscation
- Resort holidays are okay, actually
- Letting tasks linger
- My headphone collection
- My vegan cake recipe
- Becoming a runner
- My overnight oat protocol
- On being "busy"
2025
- How I fixed my sleep
- Difficulties in evaluating a deception detector for AIs
- GDM mechanistic interpretability team updates
- How to save 1/3 off TFL rail fares
- Coaching is good, actually
- Should you spend time making things more efficient?
- An opinionated guide to building a good to-do system
- everything2prompt
- My health dashboard
- Joining Google DeepMind
- Detecting strategic deception using linear probes
- Open problems in mechanistic interpretability
- Intellectual progress in 2024
- Activation space interpretability may be doomed
2024
- Book Summary: Zero to One
- Reasons for and against working on technical AI safety at a frontier AI lab
- You should remap your caps lock key
- You should consider applying to PhDs (soon!)
- Understanding positional features in layer 0 SAEs
- Unlearning via RMU is mostly shallow
- Transformer circuit faithfulness metrics are not robust
- Me, Myself, and AI: The Situational Awareness Dataset (SAD) for LLMs
2023
2021
ai
- How transparent is DiffusionGemma (and why it matters)
- Building and evaluating model diffing agents
- Training on documents about monitoring leads to CoT obfuscation
- Difficulties in evaluating a deception detector for AIs
- GDM mechanistic interpretability team updates
- everything2prompt
- Joining Google DeepMind
- Detecting strategic deception using linear probes
- Open problems in mechanistic interpretability
- Activation space interpretability may be doomed
- Reasons for and against working on technical AI safety at a frontier AI lab
- You should consider applying to PhDs (soon!)
- Understanding positional features in layer 0 SAEs
- Unlearning via RMU is mostly shallow
- Transformer circuit faithfulness metrics are not robust
- Me, Myself, and AI: The Situational Awareness Dataset (SAD) for LLMs
interpretability
- How transparent is DiffusionGemma (and why it matters)
- Building and evaluating model diffing agents
- Difficulties in evaluating a deception detector for AIs
- GDM mechanistic interpretability team updates
- Joining Google DeepMind
- Detecting strategic deception using linear probes
- Open problems in mechanistic interpretability
- Activation space interpretability may be doomed
- Understanding positional features in layer 0 SAEs
- Unlearning via RMU is mostly shallow
- Transformer circuit faithfulness metrics are not robust
productivity
- I'm organized because I'm forgetful
- Letting tasks linger
- On being "busy"
- How I fixed my sleep
- How to save 1/3 off TFL rail fares
- Coaching is good, actually
- Should you spend time making things more efficient?
- Product recommendations
- An opinionated guide to building a good to-do system
- everything2prompt
- You should remap your caps lock key
research
- How transparent is DiffusionGemma (and why it matters)
- Building and evaluating model diffing agents
- Training on documents about monitoring leads to CoT obfuscation
- Activation space interpretability may be doomed
- Understanding positional features in layer 0 SAEs
- Unlearning via RMU is mostly shallow
- Transformer circuit faithfulness metrics are not robust
- Me, Myself, and AI: The Situational Awareness Dataset (SAD) for LLMs
paper
- How transparent is DiffusionGemma (and why it matters)
- Training on documents about monitoring leads to CoT obfuscation
- Difficulties in evaluating a deception detector for AIs
- Detecting strategic deception using linear probes
- Open problems in mechanistic interpretability
- Transformer circuit faithfulness metrics are not robust
- Me, Myself, and AI: The Situational Awareness Dataset (SAD) for LLMs
personal
- I'm organized because I'm forgetful
- Photo journal
- Chesterton's preferences
- Resort holidays are okay, actually
- My headphone collection
- Becoming a runner
- Intellectual progress in 2024
careers
- Joining Google DeepMind
- Reasons for and against working on technical AI safety at a frontier AI lab
- You should consider applying to PhDs (soon!)
health
sport
books
food
physics
reading
startups
running