Home / About AI Safety

Understanding AI Safety

What is AI Safety?

AI Safety is a research field focused on ensuring that advanced artificial intelligence systems remain beneficial, aligned with human values, and under human control as they become more capable. It encompasses technical research areas like alignment, interpretability, and robustness, as well as governance considerations about how AI systems should be developed and deployed.

Why It Matters

As AI systems become more powerful and autonomous, they may develop capabilities that could lead to unintended consequences if not properly designed and controlled. The stakes are high: advanced AI could help solve humanity's greatest challenges, but also poses significant risks if developed without adequate safety measures. The field aims to maximize the benefits while minimizing potential harms.

Key Risks & Challenges

Alignment Problem
Ensuring AI systems pursue goals aligned with human values and intentions, even as they become more capable.
Interpretability
Developing techniques to understand how AI systems make decisions and represent knowledge.
Robustness
Creating systems that behave safely even when deployed in new environments or facing unexpected situations.
Power-seeking Behavior
Preventing AI systems from developing instrumental goals that conflict with human welfare.
Coordination Challenges
Ensuring that safety standards are maintained across all major AI development efforts globally.

Our Approach

Focus Areas

At BAISH - Buenos Aires AI Safety Hub, we focus on several key areas within AI safety research:

Chain of Thought interpretability
LLM evaluations
Mechanistic interpretability of neural networks

Our Contribution

We contribute to the field through:

Supporting student research projects
Building a regional community of AI safety researchers
Organizing workshops, training programs, and hackathons
Mentoring students interested in AI safety careers

Cofounders

Eitan Sprejer

Co-founding Director

Eitan is a full-time AI Safety researcher through the AISAR Scholarships program and the Apart Lab Fellowship. He has first authored AI Safety papers accepted to NeurIPS workshops and co-authored the paper "Strategic Deception in AI Systems". He holds a bachelor's + master's degree in Physics from the University of Buenos Aires. He has served as a teaching assistant at ML4G camps on two occasions and has been a facilitator for the AISES course from the Center for AI Safety. He is currently a facilitator for Blue Dot Research's AGI Strategy course.

Luca De Leo

Co-founding Director

Luca has been involved in AI Safety since 2016. After dropping out of a master's in Computer Science at the University of Buenos Aires, he received grants from ACX+ and the Long Term Future Fund to upskill for AI Safety research. He has worked in operations at Nonlinear and achieved first place in two separate Apart Research sprints and one second place. He currently does part-time operations for the AI Species YouTube channel, which has garnered over 14 million views raising awareness about AI Safety. He is currently a facilitator for Blue Dot Research's AGI Strategy course.