RESEARCH OVERVIEW
Currently, my Kensho lab's research mostly concerns LLMs and includes:
- tokenization (e.g., optimizing speed and quality; improving anisotropy; analyzing numerical biases)
- document inconsistencies (e.g., automatically detecting intra-doc issues)
- LLM evaluation (e.g., developing challenging benchmarks; analyzing LLM-Judges; radically re-thinking evaluations altogether)
- post-training methods (e.g., improving instruction-following).
If you are interested in working together at Kensho, please see our current openings. Research Scientist roles are for my lab; all other positions report to other teams. At MIT, I am not accepting grad students, and I tend to only advise 1 Master's Thesis from Harvard or MIT each year.
STUDENTS
NOTE: Most of my formal advising was during my time at Harvard (2019-2022). While at Kensho+MIT, I still collaborate with many students but due to time constraints, I rarely serve as a formal thesis adviser.
CURRENT
Joshua Wong (currently Harvard Master’s)
PAST
Haoran Zhang (graduated from Harvard Master’s)
Xiaohan Yang (Harvard Master’s Thesis 2022 -> Apple)
Anita Mahinpei (Harvard Master’s Thesis 2022)
Xin Zeng (Harvard Master’s Thesis 2022)
Jack Scudder (Harvard Master’s Thesis 2022 -> West Point Instructor)
Xavier Evans (Harvard Undergrad Independent Study)
Ning Hua (Smith x Harvard Independent Study)
Jie Sun (Harvard Independent Study -> Co-founded basys.ai)
Yoel Zweig (ETH Zurich Master’s Thesis ‘21)
Ali Hindy (High School -> Stanford CS)
Thomas Fouts (High School -> University of Michigan ME)
Mingyue Wei (Harvard Master's 2021 -> Amazon)
Alessandro Stolfo (ETH-Zurich Master’s ‘21 -> PhD program)
Brendan Falk (Harvard ‘20 -> CEO @ Fig)
EXPERIENCE
During my career within academia, industry, and government, significant projects (1-5 years) have concerned:
training LLMs (upwards of 30B params)
long-document QA
tokenization
LLM calibration
post-training
developing evaluation benchmarks
coreference resolution
sign language classification
natural language understanding (NLU)
entity linking
citation prediction
face recognition
topic modelling
machine translation
streaming algorithms for NLP
anomaly detection
adaptive web personalization
speech recognition via active learning
error-correcting codes
social network analysis
2D pattern recognition
animats-based learning (swarm intelligence)
INVITED TALKS
2024
Nov 8 — Eugene Charniak Academic Memorial/Tribute (Eugene passed June 2023)
2022-2023
Mostly internal presentations to C-levels, Executive Committees, and Board of Directors at S&P Global and Kensho.
2021
September 23 — Brown University Alumni Panel: Teaching-Track Career options
May 14 — Deep Learning with Attention @ Keystone Strategy AI/ML Speaker Series
April 30 — Hard NLP Tasks @ Harvard IACS Seminar Series [vid][slides]
January 20 — Language Models to Transformers @ Harvard ComputeFest
2020
November 20 — Research Talk @ Florida Institute of Tech.
October 15 — Career Advice @ Florida Institute of Tech.
May 19 — Open Data Science Conference (ODSC)
January 23 — Sequential Data @ Harvard ComputeFest
2019
September 27 — PhD Alumni Panel @ Brown
October 27 — RDMeetsIT Panel @ MIT Media Lab + Mercedes Benz
March 11 — Coreference Resolution @ Invitae
April 1 — MIT
March 15 — University of Washington
March 6 — CMU
February 21 — Brown
February 15 — Harvard