Guillaume Lample

Who he is

Guillaume Lample is a co-founder of Mistral AI and serves as its Chief Science Officer^[1]. The company's about page names three co-founders: Arthur Mensch as CEO, Timothée Lacroix as CTO, and Lample alongside them^[1]. Of those three titles, his is the one that points at research — a chief science officer is the person a company expects to set its scientific and technical direction, rather than run the business or the engineering organisation.

Before Mistral: the LLaMA paper

Before helping to found Mistral, Lample worked at Meta; co-founder Timothée Lacroix did as well^[4]. Lample is listed as an author of "LLaMA: Open and Efficient Foundation Language Models", posted to arXiv on 27 February 2023^[2]. The paper lists 14 authors, and Lample is the last of them^[2].

Where a name sits on a machine-learning author list carries meaning, even though the convention is rarely written down. The first name is usually whoever did most of the hands-on work; the last is usually the senior researcher who set the direction or supervised the project. Being the closing name on a 14-author paper therefore says something about the role rather than about how much code was written. arXiv, where the paper was posted, is a preprint server: papers go up there before or without formal journal publication, which is how results in this field travel quickly.

The title states the subject plainly: foundation language models, described as open and efficient^[2]. A foundation model is trained on broad data and then adapted to many tasks, rather than built for one narrow job; "efficient" points at the cost of training and running such models, which is a large part of whether other people can use them at all.

Founding Mistral AI

Mistral AI, in the company's own account, was born in April 2023 with the aim of putting frontier AI in everyone's hands^[1]. Lample is one of its co-founders and its Chief Science Officer^[1]. The distance between the LLaMA paper on 27 February 2023 and the founding in April 2023 is only a few months — the gap between publishing work on foundation models and building a company around them was short^[2][1].

Mistral 7B

Lample co-authored the Mistral 7B paper, posted to arXiv on 10 October 2023^[3]. The "7B" in the title follows a common shorthand in which the number gives a model's parameter count in billions; model papers in this field routinely carry their size in the name so that readers know what class of system is being described before reading further^[3].

How the record reads

Put the dated markers in order and you get a compact sequence: the LLaMA paper on 27 February 2023, the founding of Mistral AI in April 2023, and the Mistral 7B paper on 10 October 2023^[2][1][3]. All three sit in the same calendar year, and the same person's name appears at each stage — as an author of a foundation-model paper, as a co-founder of a company that says it wants to put frontier AI in everyone's hands, and as an author of the company's model paper^[2][1][3].

That is the shape of the public record of his work: research on language models, then a company built around them, with him holding the science title in a three-person founding team whose other two roles are the executive and technical ones^[1].