Myle Ott

Myle Ott is an AI researcher whose work spans deceptive-review detection, machine translation and language-model training at scale. At Cornell, he presented 2011 software that identified 90 percent of deceptive reviews among 800 Chicago hotel reviews, then coauthored a 2013 NAACL paper on opinion spam. He led the author lists for neural-translation and fairseq papers, coauthored RoBERTa and OPT, helped introduce FSDP, and his deceptive-review paper received the ACL's 2021 Test-of-Time Paper Award. His 2025 work with Thinking Machines includes Tinker, a flexible API for fine-tuning language models. As of October 2026, he is part of Thinking Machines' founding team; earlier, he served on Character AI's founding team and led large language model efforts at Meta AI (FAIR).

Now at: Thinking Machines Lab
Works: 65
Badges: Founder

Publication numbers from OpenAlex, as of 22 October 2026. Citations and h-index are hidden because this OpenAlex record may be incomplete or mix in other authors.

Timeline

2011–2013

Opinion-spam research at Cornell

In 2011, while a Cornell doctoral candidate in computer science, he presents the opinion-spam study at the Association for Computational Linguistics meeting in Portland; software picks out 90 percent of deceptive reviews in 800 Chicago hotel reviews. In 2013, he coauthors “Negative Deceptive Opinion Spam” with Claire Cardie and Jeffrey T. Hancock in the NAACL proceedings.
2018

Scaling neural machine translation

In October 2018, he leads the author list of “Scaling Neural Machine Translation” with Sergey Edunov, David Grangier and Michael Auli. The paper shows that reduced precision and large-batch training speed up training by nearly 5x on a single 8-GPU machine, and reaches 29.3 BLEU on WMT'14 English-German after 85 minutes on 128 GPUs.
2019

The fairseq toolkit and RoBERTa

At Facebook AI Research, he leads the author list of the 2019 NAACL demonstrations paper for fairseq, an open-source toolkit for translation, summarization and language modeling. In July 2019, he coauthors RoBERTa, a replication study of BERT pretraining that finds BERT significantly undertrained.
July 2021

Fully Sharded Data Parallel

In July 2021, he leads the author list of a Facebook AI Research Engineering post introducing Fully Sharded Data Parallel (FSDP). The method shards model parameters, gradients and optimizer states across GPUs; it enables training models orders of magnitude larger with fewer GPUs.
2021

ACL Test-of-Time Paper Award

In 2021, the ACL gives the Test-of-Time Paper Award for 2011 to “Finding Deceptive Opinion Spam by Any Stretch of the Imagination,” which he writes with Yejin Choi, Claire Cardie and Jeff Hancock.
May 2022

Open Pre-trained Transformers at Meta AI

In May 2022, he is one of the 19 authors of Meta AI's Open Pre-trained Transformers (OPT), a suite of decoder-only language models from 125M to 175B parameters that the team aims to share with researchers.
2025

Tinker fine-tuning API

In 2025, he works with the Thinking Machines team on Tinker, a flexible API for fine-tuning language models.
As of October 2026

Thinking Machines founding team

As of October 2026, he is part of the founding team at Thinking Machines. Earlier, he serves on Character AI's founding team and leads large language model efforts at Meta AI (FAIR).