engblogs

A remarkable journey from Laredo, Texas, to leading the MIT autonomous vehicle team and earning an MBA from MIT Sloan.

Training a model to improve mathematical reasoning by rewarding correct steps of reasoning instead of just the final answer, leading to better performance and aligning the model with human thinking.

Large sequence models for software development activities that use the process of software development as the source of training data, resulting in promising results for professional software developers and the potential for developing agents that can assist across the software development process.

A glimpse into the daily responsibilities and projects of a senior software engineer on the Photo Publishing Team at the New York Times.

Snowflake Snowpark enables organizations to take ML experiments into production by providing production-grade ML data pipelines, data preparation in Python, and leveraging Scikit-Learn for feature engineering and ML model training in Snowflake.

Comparing the Rust and OCaml programming languages' system for tracking lifetime and ownership

Reflections on the impact of the Institute for Data, Systems, and Society at MIT during a two-day conference, ahead of the departure of its founding Director Munther Dahleh.

Introducing the Barkour agility benchmark for quadruped robots, which measures robot agility and mobility through a diverse and challenging obstacle course, encouraging the development of fast and versatile locomotion controllers.

This blogpost discusses the challenges of reasoning on charts in visual language and proposes a foundation model called MatCha, which is trained on chart de-rendering and math reasoning tasks, surpassing previous state-of-the-art models in ChartQA.

How to incorporate democratic inputs into AI development

Exploring the history, capabilities, and limitations of large language models, from the emergence of ChatGPT to the evolution of models like BERT and GPT-3, discussing key terms like attention, embeddings, and transformers, along with ways to adapt and use these models for specific tasks.

A research framework for evaluating general-purpose AI models against novel threats, including extreme risks, to ensure responsible development and deployment.