Data Science Group @BCAM | Codex as a Research Assistant: Tools, Models, and Limitations

​In this talk, I will give an informal introduction to how Codex works, including tasks, agents and subagents, goals, tools, and the typical plan–execute–steer–verify workflow. I will then discuss how these features can support machine-learning research, focusing on model selection, context and token efficiency, literature search, hallucinations, and verification. A short example will illustrate a research-oriented coding task. Participation is encouraged, and attendees are invited to share their own workflows, failures, and tips.