Michael C. Mozer

What if a foundation model could tell us how to modify its architecture to boost inference and reasoning instantaneously—without retraining? What if that tweak incurred near zero latency cost during generation and supported indefinite state tracking? arxiv.org/abs/2608.17981

[1/5] Intelligent behavior relies on maintaining an evolving, dynamic model of the environment. But even frontier models sometimes fail at tracking state in dialogs, e.g., actual transcript from 4/20/2026:

[1/4] As you read words in this text, your brain adjusts fixation durations to facilitate comprehension. Inspired by human reading behavior, we propose a supervised objective that trains an LLM to dynamically determine the number of compute steps for each input token.

Overcoming temptation: Incentive design for intertemporal choice
arxiv.org/abs/2203.05782
We use AI models to help individuals adhere to long-term goals (e.g., retirement savings, weight loss) and avoid giving in to temptation.

My first tweet