coby.lk
CVGitHubLinkedInEmail

I'm Coby, and I'm an AI safety researcher with interest in control and science of generalization. I have a broader interest in most things related to macrostrategy, existential risk, analytical philosophy, and EA/AIS fieldbuilding. In my free time, I love composing and performing music, dancing Argentine tango, and (ski) mountaineering. I'm currently doing my undergrad at Yale, where I help organize Yale Effective Altruism and Yale AI Alignment.

NEWS
DRAG →
June 2026
I'm now working on the Tracks Team at the UChicago Existential Risk Laboratory, creating Khan-Academy-shaped courses for AI safety.
February 2026
We've started a research project having to do with emergent misalignment.
January 2026
I'm now head of operations for Yale AI Alignment and the fellowship manager for YEA.
December 2025
We finished our privacy neuron-editing project, comparing inference-time interventions for suppressing PII leakage from LLMs.
October 2025
I joined the board of Yale Effective Altruism (YEA), and I'm a facilitator for our intro fellowship.
August 2025
We wrapped up extensions of our SPAR project, producing a manifold neural block with differentiable Betti numbers and a simplex-constrained transformer.
May 2025
I finished SPAR and presented our findings so far at the poster session. We're continuing the project beyond the program.
February 2025
I was accepted to the SPAR program under Dr. Ronak Mehta to research models that are inherently interpretable.
November 2024
Our activation-steering data-extraction project placed 7th in the LLM Privacy Contest at NeurIPS.
May 2016
I graduated the third grade; I could not have achieved this momentous milestone without the support of my beloved family and friends.
February 2009
I am enormously excited to share that I have just turned two years old.
SELECTED RESEARCH
Privacy Neuron Editing2025Manifold Neural Block2025Simplex-Constrained Transformer2025SPAR: Simplicial Neural Networks2025CAA Data Extraction2024
WRITING
lasso, smoothing, and kernels2026kernels and neural networks2026

Coby Kassner