Todor Markov
LinkedIn
Github
Bio
- 2024+ Researcher at Anthropic, Knowledge team. Teaching Claude to use retrieval tools.
- 2023-2024: Researcher at OpenAI, Preparedness team. Worked on evaluating the persuasion abilities of frontier AI systems
- 2023 (May) - 2023 (November): Researcher at OpenAI, Safety Systems team. Worked on automatic content moderation for the OpenAI API and ChatGPT
- 2020 - 2023 (May): Researcher at OpenAI, Applied AI team. Worked on fine-tuning and content moderation for the OpenAI API
- 2018 - 2020: Researcher at OpenAI, Multi-Agent team. Worked on multi-agent deep reinforcement learning in hide-and-seek
- 2017 - 2018: Backend software engineer at Blend
- 2013 - 2017: Stanford B.S. in Symbolic Systems, M.S. in Statistics. Research on using machine learning to discover fast-charging protocols for batteries, published in Nature
Core values and beliefs
I care deeply about advancing freedom, both for myself and for others. To me, that means:
- Creating a world where no person is forced to choose between sacrificing their own health or happiness, or not being able to provide for themselves and their families
- Helping people better understand what their core wants and needs are. Giving people better mental & emotional tools for resolving conflicting internal desires
- Advancing our understanding of the world itself, the conseqences of our actions, and the consequences of our wants
On a more concrete level, I want to see progress in the following areas:
-
Overall welfare and prosperity:
- General quality of life: health-adjusted life expectancy; life satisfaction; material prosperity; free time
- Education & access to information. Empowering people to make informed choices about their life
- Civil liberties, rule of law and nondiscrimination
- Reducing global catastrophic risks like climate change, nuclear war, large-scale pandemics and potential risks from artificial intelligence
-
Designing institutions, processes and tools that effectively support the above two areas
- Improving the full science & technological development cycle, from fundamental research to commercialization and deployment
- Creating better governance structures
- Reforming the education system
I believe that scientific and technological advancements are crucial for creating the kind of progress I describe above (see these two blog posts for some specific arguments behind this position)
My current work ultimately aims to create human-level artificial intelligence that is aligned with human values, and ensure that the benefits of that technology are distributed broadly across the whole of humanity. I believe that the potential impact of this work is extremely high - if we are successful, that would lead to massive progress in all the areas I describe above. Conversely, if sufficiently powerful AI systems are misused or deployed in an unsafe manner, that could cause massive, potentially irreversible damage.
Personal
I was born and raised in Bulgaria. There, I studied at a gymnasium that focused on mathematics and natural sciences, and I was very active in math contests (culminating in competing at the International Mathematical Olympiad in 2012 and 2013).
In my free time, I like spending time with friends, reading, travel and adventure sports.
Writing
The below documents are compressed summaries of my thoughts on a given topic, synthesized via reading a number of books and other resources. Note that they were originally written for my personal use, so they might be missing important implicit context. This could potentially make them difficult to understand for external readers.
Creating, modifying and eliminating habits
Effective learning
Emotions and mental state management
Project management
Publications
-
Building an early warning system for LLM-aided biological threat creation
Tejal Patwardhan, Kevin Liu, Todor Markov, Neil Chowdhury, Dillon Leet, Natalie Cone, Caitlin Maltbie, Joost Huizinga, Carroll Wainwright, Shawn (Froggi) Jackson, Steven Adler, Rocco Casagrande, Aleksander Madry
[blog]
-
A Holistic Approach to Undesired Content Detection in the Real World
Todor Markov, Chong Zhang, Sandhini Agarwal, Tyna Eloundou, Teddy Lee, Steven Adler, Angela Jiang, Lilian Weng.
Oral presentation at the Thirty-Seventh AAAI Conference on Artificial Intelligence (AAAI), 2023
[paper][blog]
-
Autonomous Screening and Optimization of Battery Formation and Cycling Procedures
Stefano Ermon, William C. Chueh, Aditya Grover, Todor Markov, Nicholas Perkins, Peter M. Attia
United States Patent. Patent No.: US 10,992,156 B2. April 2021
[patent]
-
Closed-loop optimization of fast-charging protocols for batteries with machine learning
Peter Attia, Aditya Grover, Norman Jin, Kirsten Severson, Todor Markov, Yang-Hung Liao, Michael Chen, Bryan Cheong, Nicholas Perkins, Zi Yang, Patrick Herring, Muratahan Aykol, Stephen Harris, Richard Braatz, Stefano Ermon, William Chueh
Nature, February 2020.
[paper]
-
Emergent Tool Use From Multi-Agent Autocurricula
Bowen Baker, Ingmar Kanitscheider, Todor Markov, Yi Wu, Glenn Powell, Bob McGrew, Igor Mordatch
Spotlight talk at International Conference on Learning Representations (ICLR), 2020.
[paper][code][blog]
-
Best arm identification in multi-armed bandits with delayed and partial feedback
Aditya Grover, Todor Markov, Norman Jin, Peter Attia, Nick Perkins, Bryan Cheong, Michael Chen, Zi Yang, Stephen Harris, William Chueh, Stefano Ermon
International Conference on Artificial Intelligence and Statistics (AISTATS), 2018.
[paper][code]