Too much of a good thing: how context can affect answers from AI

For anyone who uses AI on models, drawings and regulations. You learn what the context is, why everything is sent again at every step of a conversation or an agent, and why more context can give worse answers. In a test with an IFC model and the Norwegian building code TEK17, the AI model approved a fire compartment that is missing a wall, in all ten runs, when it was given the whole building and the rules. Given only the room, it found the gap every time. The video ends with how to tidy the context, both by hand and with agents.

The recording, subtitles and member summary remain in their original language.

Watch the full video as a member

Employees of active member companies gain access to the recording and the professional summary.