about me¶

Chris Merck

I'm an entrepreneur-engineer, researcher, and linguaphile. I am co-founder and CTO at Bond Home, where I lead a team building smart home experiences.

Someday I shall build delightful language learning software, so that anyone may benefit from the wonders of comprehensable input and spaced repetition, all while having fun and enhancing international goodwill.

In the meantime, we have our work cut out for us ensuring advances in artificial intelligence are aligned with human interests. I'm encouraged by the successes of Constitutional Alignment, but we have yet to really grapple with the challenges of multi-agent reinforcement learning and more generally the coupling of powerful agentic AI with the economy.

Looking for collaborations with researchers and safety-focused organizations in the areas of scalable alignment and interpretability. As Nick Bostrom says:

the most appropriate attitude may be a bitter determination to be as competent as we can, mack as if we were preparing for a difficult exam that will either realize our dreams or obliterate them

Bring it on!

experiences¶

I have been reading interpretability papers and trying to reproduce their figures.

I received a small grant to investigate reasoning failures in DeepSeek models.

I built a home control demo the morning that gpt-4o was released.

I've been dabbling with NLP/ML for a long while, origially inspired by Manning & Schuetze. I'm a fan of spacial visualizations (2011) and multimodality (2013).

I studied physics and computer science at Stevens, writing a wave tank controller for NASA Wallops, video annotation software to label data for an NIH-funded eating study, and channeled Russel Monroe whenever making figures.

Kleinberg Lab

blog highlights¶

Humpty Dumpty Successfully Annealed (2026) — Solving a Jane Street puzzle: reassembling a neural network's scrambled layers with the Hungarian algorithm and SVD
Considerations when Building Embedded Databases (2024) — Pitfalls of persisting device state to flash memory, originally published on Memfault's Interrupt blog
Clustering Jane Austen (2011) — Hierarchical clustering on word bigrams across Austen's novels, with dendrograms and word clouds
Easy Syntax Trees (2011) — An informal diagramming technique applied to English, Chinese, and a long Kafka sentence
Korean Kuriosities (2011) — Observations on Hanja, borrowed words, and calligraphy from a trip to Seoul
Extending the Knuth Operator (2011) — Exploring fractional extensions of Knuth's up-arrow notation
Option Pricing and Brownian Motion (2011) — Connecting stock option pricing theory to the physics of diffusion
Playing Genetic Doublets in 3D (2010) — Lewis Carroll's word ladder game solved with genetic algorithms and embedded in 3D

selected papers¶

Causal Explaination, AAAI (2016) - Computing the strength of causal relationships between events in a physical model.
Physical Model Theory (2010) - A broad exploration of the simplest world-models used in physics.
Fuzzy Flip-Flops (2010) - Relaxes the binary nature of logic gates to explore continuous-valued logic.
Finite Machines (2009) - A formal exploration of small machines with limited memory.
Sunglint Statistics, Applied Optics (2008) - We study the pattern of light reflected from the ocean surface in order to help remove it from satellite imagery. I'm most proud of the attractor diagrams that reveal structure in the glint time series.

misc¶

Otherwise I spend my time playing guitar, taking friends foraging, ever so gradually learning Latin, and teaching kids to solder and code. My intent with this site is to share occasional writeups related to work, research, and adjacent.