Can large AI models acquire key features of human intelligence? Invitation to a lecture by Loïc Barrault from Meta AI

A major international research workshop, JSALT 2025, organized by Johns Hopkins University (Maryland, USA), is taking place at the Faculty of Information Technology at BUT and will run until the end of July. Thanks to this event, around one hundred leading global researchers in language and speech technologies have come to Brno. Four teams are working for six weeks on current research topics related, for example, to large language models or speech processing under challenging conditions. The workshop program is further enriched by plenary lectures delivered by prominent experts from both industry and academia. An especially interesting talk is expected on Tuesday, July 15, when Loïc Barrault from Meta AI in France will give a presentation.

Loïc Barrault will speak at 11:00 at our faculty with his lecture titled Large Concept Model: beyond token-based Large Language Models. In this talk, he will explore (certainly also socially somewhat provocative) questions such as whether large AI models can acquire key characteristics of human intelligence, such as reasoning, planning, hierarchical, or multilingual information processing. The very topic invites a wider audience — you can either attend in person in room E112 at FIT BUT or follow the live stream on the JSALT YouTube channel.

It should be added that Loïc Barrault is a renowned researcher in the field of machine translation, who has worked at universities in Le Mans and Sheffield, among others. Today, he is recognized as one of the pioneers in integrating language, speech, and visual information, significantly contributing to the development of adaptive models over the long term.

You can find more information about the workshop, which can also be seen as a recognition of the long-standing work of the Brno-based BUT Speech@FIT research group and its head, Professor Jan Černocký, in the press release HERE.