Albert Jiang

Career

Albert Jiang has been a research scientist at Mistral AI since the company's formation in June 2023, and Jiang's team there works on the science and infrastructure of reasoning. Earlier, Jiang's CV lists a year as a research assistant at the University of Toronto with Jimmy Ba and Roger Grosse (2019–2020) and a research scientist internship at FAIR with Guillaume Lample in 2022.¹²

Jiang's PhD at the Cambridge Computer Laboratory, supervised by Mateja Jamnik and Wenda Li, studied how language models can learn abstract mathematical reasoning; the thesis was examined in October 2024.¹

Research

During the PhD, Jiang worked on autoformalization of theorems and proofs, premise selection tools for language models, and how humans and language models interact on mathematical tasks. Jiang is first author of Thor (2022), which integrates language models with automated theorem provers, and of "Draft, Sketch, and Prove" (2022), and a co-author of "Autoformalization with Large Language Models" (2022).¹²⁵⁶⁷

At Mistral AI, Jiang is first author of the Mistral 7B (2023) and Mixtral of Experts (2024) papers, whose models are released under the Apache 2.0 license. Jiang's homepage describes later work on mid- and post-training data for Mathstral in 2024, large-scale reinforcement learning for Magistral in 2025, and the formal agentic models Leanstral 1.0 and 1.5 in 2026.³⁴¹

Work and contributions

• Mistral 7B and Mixtral 8x7B Open weights

  First author of the Mistral 7B (2023) and Mixtral of Experts (2024) papers, whose abstracts say the models are released under the Apache 2.0 license; Jiang's homepage says this work was on pretraining data.³⁴¹

• Mathstral, Magistral and Leanstral

  Jiang's homepage lists work on mid- and post-training data for Mathstral (2024), large-scale reinforcement learning for Magistral (2025), and formal agentic models Leanstral 1.0 and 1.5 (2026) at Mistral AI.¹

• Draft, Sketch, and Prove: Guiding Formal Theorem Provers with Informal Proofs Published research

  First author of the 2022 paper on using informal proofs to guide formal theorem provers.⁶

Timeline

1. Jun 2019Began a year as a research assistant at the University of Toronto with Jimmy Ba and Roger Grosse.²
2. May 22, 2022Posted Thor, on integrating language models and automated theorem provers, as first author.⁵
3. Jul 2022Began a research scientist internship at FAIR with Guillaume Lample, through October 2022.²
4. Oct 21, 2022Posted "Draft, Sketch, and Prove" as first author.⁶
5. Jun 2023Joined Mistral AI as a research scientist at the company's formation.¹
6. Oct 10, 2023Posted the Mistral 7B paper as first author.³
7. Jan 8, 2024Posted the Mixtral of Experts paper as first author.⁴
8. Oct 2024Had the Cambridge PhD thesis on learning mathematical reasoning with language models examined.¹
9. 2025Worked on large-scale reinforcement learning for Magistral at Mistral AI.¹
10. 2026Worked on the formal agentic models Leanstral 1.0 and 1.5.¹

Other documented roles

Roles the sources document that are not confirmed as current.

• Research Scientist Intern (July to October 2022), FAIR (Meta)²

Where they work

• Mistral AI 4 people in the wiki

Related people

• Arthur MenschAlso at Mistral AI
• Guillaume LampleAlso at Mistral AI
• Timothée LacroixAlso at Mistral AI
• Alexander WettigLanguage models · Datasets and data
• Marina DanilevskyLanguage models · Datasets and data
• Weizhu ChenLanguage models · Reinforcement learning
• John SchulmanReinforcement learning · Language models
• Connor LeahyLanguage models · Datasets and data
• Ranit AharonovLanguage models · Datasets and data
• Rangan MajumderLanguage models · Datasets and data