Jason Weston

Career

Jason Weston earned a PhD in machine learning at Royal Holloway, University of London and at AT&T Research, according to Meta's people page. Weston was then a researcher at Biowulf Technologies, a research scientist at the Max Planck Institute for Biological Cybernetics, a research staff member at NEC Labs America, and a research scientist at Google.¹

Weston is now a research scientist at Meta, and Meta's page also lists a visiting research professorship at NYU. Meta's page notes that Weston was part of the YouTube team recognized with a Technology and Engineering Emmy Award for personalized recommendation engines for video discovery.¹

Research

With Ronan Collobert, Weston wrote "A Unified Architecture for Natural Language Processing: Deep Neural Networks with Multitask Learning" (ICML 2008), which later received the ICML Test of Time Award. In 2014 Weston, Sumit Chopra and Antoine Bordes introduced Memory Networks, models that pair inference components with a long-term memory that can be read and written.⁹²

Much of Weston's later work concerns dialogue: the open-source ParlAI platform (2017), the open-domain chatbot recipes paper (2020) and the deployed BlenderBot 3 agent (2022), alongside the Adversarial NLI benchmark (2019). Recent co-authored papers study language models that judge and reward their own outputs (2024), the Byte Latent Transformer, which works with patches rather than tokens (2024), and meta-reasoning controllers for long-running agents (2026).^3564718

Work and contributions

• Memory Networks Published research

  First author, with Sumit Chopra and Antoine Bordes, of Memory Networks (2014), which describes learning models that combine inference components with a long-term memory that can be read and written, studied on question answering.²

• ParlAI Open-source code

  Co-author of ParlAI: A Dialog Research Software Platform (2017), an open-source Python platform for sharing, training, and testing dialog models; the repository is published under the MIT License.³¹⁰

• Self-Rewarding Language Models Published research

  Co-author of Self-Rewarding Language Models (2024), which studies language models that provide their own rewards through LLM-as-a-Judge prompting during iterative DPO training.⁷

Timeline

1. 2008Presented a unified multitask neural architecture for natural language processing with Ronan Collobert at ICML.⁹
2. Oct 15, 2014Posted Memory Networks with Sumit Chopra and Antoine Bordes.²
3. May 18, 2017Co-authored the paper introducing ParlAI, an open-source dialog research platform.³
4. Oct 31, 2019Co-authored Adversarial NLI, a benchmark for natural language understanding.⁴
5. Apr 28, 2020Co-authored "Recipes for building an open-domain chatbot".⁵
6. Aug 5, 2022Co-authored the BlenderBot 3 paper on a deployed conversational agent that continually learns.⁶
7. Jan 18, 2024Co-authored Self-Rewarding Language Models.⁷
8. Dec 12, 2024Co-authored Byte Latent Transformer: Patches Scale Better Than Tokens.¹
9. Sep 29, 2026Co-authored a paper on agentic meta-reasoning for scaling inference in long agent runs.⁸

Other documented roles

Roles the sources document that are not confirmed as current.

• Research scientist, Google¹

Honors and recognition

• ICML Test of Time Award · International Conference on Machine Learning, 2018⁹¹
• Technology and Engineering Emmy Award (YouTube team, personalized recommendation engines for video discovery) · National Academy of Television Arts & Sciences¹

Where they work

• Meta 106 people in the wiki

Related people

• Thomas ScialomAlso at Meta
• Jason WeiAlso at Meta
• Gabe PereyraAlso at Meta
• Noam BrownAlso at Meta
• Yuxin WuAlso at Meta
• Jiaxuan YouAgents · Language models
• TekniumLanguage models · Agents
• Jacob AndreouAgents · Language models
• Yoav ShohamLanguage models · Agents
• Shunyu YaoAgents · Language models