Theo X. Olausson

Theo X. Olausson (he/him)

CV · GitHub · Google Scholar · Email

I'm a PhD candidate at MIT working with Armando Solar-Lezama on building reliable and steerable language models.

I started my PhD with early work on self-correcting code generation [OIW+24, GLJ+24], abstraction learning [BOW+23, GWB+24] and combining language models with symbolic tools [OGL+23]. More recently, I have worked on diffusion language models [JOB+26, OJW+26, BTM+26] and statistically guaranteed hallucination control [EOS+26]. In addition to MIT, I've had the good fortune of working on these topics (and more) at Apple ML Research, the Center for Computational Mathematics at the Flatiron Institute, and Microsoft Research.

I'm passionate about community building and have previously served as co-President of the MIT EECS Graduate Students' Association. I'm also one of the lead organizers of VerifAI, the Workshop on AI Verification in the Wild (held at ICLR'25 and '26). Most of my free time is spent doing some variation of cycling, running or climbing. I also like film photography, gothic novels and writing music.

News

Publications

Diffusion Language Models (3)

A Tale of Two Temperatures: Simple, Efficient, and Diverse Sampling from Diffusion Language Models

COLM'26

TX. Olausson*, M. Jazbec*, X. Wang, A. Solar-Lezama, CA. Naesseth, S. Mandt, E. Nalisnick

arXiv:2604.09921 (2026)
arXiv

The Design Space of Tri-Modal Masked Diffusion Models

In Review

L. Béthune, V. Turrisi, BK. Mlodozeniec, PR. Lopez, L. Boominathan, ..., TX. Olausson, ..., J. Ramapuram

arXiv:2602.21472 (2026)
arXiv

Learning Unmasking Policies for Diffusion Language Models

ICML'26 Oral Spotlight Best Paper @ DeLTa 2026

M. Jazbec*, TX. Olausson*, L. Béthune, P. Ablin, M. Kirchhof, J. Monteiro, V. Turrisi, J. Ramapuram, M. Cuturi

The Forty-Third International Conference on Machine Learning (ICML 2026)
arXiv

Self-Correcting Code Language Models (2)

The Counterfeit Conundrum: Can Code Language Models Grasp the Nuances of Their Incorrect Generations?

ACL'24 Findings

A. Gu, WD. Li*, N. Jain*, TX. Olausson*, C. Lee*, K. Sen, A. Solar-Lezama

Findings of the 62nd Annual Meeting of the Association for Computational Linguistics (Findings: ACL 2024)
arXiv

Is Self-Repair a Silver Bullet for Code Generation?

ICLR'24

TX. Olausson, JP. Inala, C. Wang, J. Gao, A. Solar-Lezama

The Twelfth International Conference on Learning Representations (ICLR 2024)
arXiv

Tool Use & Discovery (3)

LILO: Learning Interpretable Libraries by Compressing and Documenting Code

ICLR'24

G. Grand, L. Wong, M. Bowers, TX. Olausson, JB. Tenenbaum, J. Andreas

The Twelfth International Conference on Learning Representations (ICLR 2024)
arXiv

LINC: A Neuro-Symbolic Approach for Logical Reasoning by Combining Language Models with First-Order Logic Provers

EMNLP'23 Outstanding Paper

TX. Olausson*, A. Gu*, B. Lipkin*, C. Zhang*, A. Solar-Lezama, JB. Tenenbaum, R. Levy

The 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP 2023)
arXiv

Top-Down Synthesis For Library Learning

POPL'23

M. Bowers, TX. Olausson, L. Wong, G. Grand, JB. Tenenbaum, K. Ellis, A. Solar-Lezama

The 50th ACM SIGPLAN Symposium on Principles of Programming Languages (POPL 2023)
arXiv

Miscellaneous (2)

Amortizing Maximum Inner Product Search with Learned Support Functions

ICML'26

TX. Olausson, J. Monteiro, M. Klein, M. Cuturi

The Forty-Third International Conference on Machine Learning (ICML 2026)
arXiv

HeteroGen: Automatic Synthesis of Heterogeneous Cache Coherence Protocols

HPCA'22 IEEE Micro Top Picks

N. Oswald, V. Gavrielatos, V. Nagarajan, T. Olausson, DJ. Sorin, R. Carr

The 28th IEEE International Symposium on High-Performance Computer Architecture (HPCA 2022)