Theo Browne discusses reported researcher departures and his view of Google’s approach to coding models. Theo Browne distinguishes model knowledge from practical performance on long-running agent tasksEvaluation is the systematic assessment of a system's behaviour or results against explicit criteria..
Theo Browne demonstrates a coding assistant examining a project and comments on its reasoning and tool use. Theo Browne also describes why he thinks development histories and feedbackAn AI feedback loop occurs when AI outputs or their consequences become inputs that influence later model or agent behavior. may matter for training useful coding agentsCode generation uses AI or another automated system to create source code from instructions, examples, schemas, or higher-level specifications..
Theo Browne discusses an engineer’s public account of dismissal after creating a Workspace command-line tool and compares it with his understanding of experiments at other AI companies. These explanations and conclusions are Theo Browne’s interpretations.
Watch on YouTube
