AI Capabilities Vocabulary

capabilities that suddenly appear at certain model scales without being explicitly trained

“Chain-of-thought reasoning emerged only in models above a certain size.”

learning to perform new tasks from examples provided in the prompt without weight updates

“In-context learning let the model translate to a new language from just five examples.”

applying knowledge learned from one task to perform better on different tasks

“Transfer learning from English improved the model's performance on French.”

capable of processing multiple types of input like text, images, and audio

“The multimodal model could answer questions about uploaded photos.”

the ability to draw conclusions through logical steps from given information

“Multi-step reasoning allowed the model to solve complex word problems.”

an internal representation of how the world works used for prediction and planning

“The model's world model understood that dropped objects fall down.”

building complex meanings from combinations of simpler parts

“Compositionality lets us understand 'blue rectangular box' from knowing each word.”

applying learned patterns to new, previously unseen situations

“Good generalization meant the model could classify animals it never saw in training.”

forming general concepts from specific instances

“The model learned the abstraction 'mammal' from examples of dogs, cats, and whales.”

connecting language to real-world entities, actions, or perceptions

“Tool use provides grounding by letting the model verify facts against reality.”