"The state of things is evidently far from simple" - William James
Latest highlighted read:
Things I worked on
Fieldbuilding, leading the policy efforts in AI to ensure a positive future.
learn more →
Working with Stepan Shabalin to apply SAEs for mech interp on VLAs.
A paper analyzing US legislation trends around deepfakes in the 2020-2025 time period, produced as my final artifact for INTA8000 under Prof. Margaret Kosal.
read here →
A published Google Docs extension for selective word counting. 2K+ downloads!
website + download →Sources I recommend
Mostly policy, love their weekly briefings!
visit source →More technical, great analysis of model cards.
visit source →More anthropological/social commentary shaped.
visit source →Exactly what you'd assume from the title.
visit source →Great books
An absurdist comedy all about self-forgiveness with outstandingly well-written female characters. Probably my favorite work of literature ever, the dialogue is both histerical and poignant.
The prime example of "esperpento" a Spanish play genre that can be understood as being a tragedy with strong social critique. Beautiful but soul-crushing, I was in a funk for a few days.
Ted Chiang is a favorite between sci-fi fans and this short story along with "Hell is the Absence of God" are my favorite.
Fantastic long-form essay-argument, made me grateful for the randomness inherent in human talent and encouraged my first public Substack
It's up there with Fahrenheit 451, but I found the concept of SOMA more troubling. Weirdly, I did not enjoy 1984 as much.
Very interesting reflections on the role of the law. Reading it was the first time I encountered the argument that suing and leveraging the law as the civilized alternative to direct violent acts is merely a change in the form of violence. The one suing simply defers their desire for retribution to the law. You can find the PDF here.
Hey, I'm Rocio! B.S./M.S. Comp Sci @ Georgia Tech, minor in Law, Science and Technology. think of the c as a th and you're good to go.
I want to understand what good AI Policy actually looks like so I can be useful in shaping it. I co-direct the AI Safety Initiative at Georgia Tech, leading field-building efforts to get the very best working on preventing harms from transformative AI. Check out some of my thoughts on this on Substack.
What I'm up to
- IAPS summer fellowship, working on policy research to secure a positive world with advanced AI
- Co-directing AISI, read about us here
- Interpretability research on VLAs in the PAIR Lab
- Wrapped up: AI@Academics internal strategy initiative under Provost Raheem Beyah
If you're at Tech and want to get involved with AISI, come to our next event and feel free to reach out
Provide anonymous feedback here
rocio.perales (at) gatech.edu
After 10pm on Thursday you can find me at a concert, out with friends, re-reading Physical Intelligence's papers or at the gym.
Education
Skills
Professional & Research Experience
- Conducting robotics threat modeling research on the Frontier Security team
- Synthesized the 269 Great American AI Act discussion draft into a 2-pager to inform the Policy Team's decision-making
- Produced an explainer of the FCC preventing foreign new advanced robotics from receiving equipment authorization
- Conducting robotics interpretability research with Prof. Animesh Garg, applying Sparse Autoencoders (SAEs) to vision-language-action models with guidance from Stepan Shabalin
- Trained YOLO models on AWS-based GPU instances for electrical infrastructure anomaly detection
- Integrated convolutional autoencoders to detect anomalies in powerline inspections for compliance detection
Leadership & Campus Involvement
- Selected by Provost Raheem Beyah to chair subcommittee for Georgia Tech's institutional AI strategy
- Designed and deployed campus-wide survey collecting 1K+ student responses on AI usage and concerns
- Results will inform GT's strategic AI plan by May 2026
- Lead 15-person exec organization and oversee task delegation for AI safety research and education
- Manage $30K+ of external funding and our Policy Fellowship curriculum and Fellowship
- Led deepfake governance research team under Dr. Margaret Kosal; invited to Anthropic Economic Futures Symposium
- Previously Communications and Media lead (Sept. 2025)
- Contributed to large-scale robotics software framework (C++/Linux/ROS2)
- Led technical challenge implementation resulting in 2nd place and 4th overall at RoboCup 2025 Intl
- Contributed to multi-robot pathfinding system with real-time decision-making under uncertainty