A team from Carnegie Mellon has developed a system called LegoGPT that enables users to construct physical LEGO models using text prompts.
The team emphasises that past innovations have focussed on generating 3D shapes, but their aim is to use AI to transform text into actual LEGO constructions.
They do this by training data models using one of the LLaMA models, which involves converting Lego designs into tokens, similar to a chatbot.
The process then renders the design into a mesh before converting it into brick sizes that are stable, to be finally rendered into a picture and sent to a robot for it to be physically built.
The team hopes to build more complex designs in the future, such as the Turing machine.
