The AI Field Guideby Stephen Hunt's SFcrowsnest
The AI Field Guide / W
3 terms, explained without the techno-murk.
AI designed for a limited task or range of tasks rather than general human-like ability.
'Weak' does not mean useless or poor. A narrow system may outperform every person at one job while being unable to do anything outside it, like a brilliant chess player who cannot make a cup of tea.
A model that detects tumours in scans but cannot write an email is narrow AI.
The learned numbers inside a neural network that control how strongly signals affect one another.
Weights are a major part of a model's parameters. Training adjusts them across many examples. A downloadable model is often distributed as a large weights file plus code and configuration.
Tiny changes across many weights gradually improve a model's predictions.
An internal model of how an environment works and how it may change after an action.
It is like a small simulator inside the AI. By imagining possible next states, the system can plan before acting. A world model is always a simplified representation and may miss important parts of reality.
A robot predicts that pushing a cup near the table's edge could make it fall.