In this session, a broad range of topics was discussed, starting from Beliz Günel’s career journey and extending to the pre-training and post-training processes of Gemini models, dataset and task design approaches for enhancing LLM capabilities, and the evaluation methods used to assess these tasks.

The session addressed the development and training processes of large language models, efforts to introduce new capabilities, methodologies for measuring model performance, the attention challenge commonly referred to as “Lost in the Middle,” the working culture within the Google DeepMind team, and the importance of an academic career path in a rapidly evolving technological landscape.