The AI Field Guideby Stephen Hunt's SFcrowsnest
The AI Field Guide / O
4 terms, explained without the techno-murk.
A model released with some parts available for others to inspect, use or adapt.
'Open' can mean different things: downloadable weights, open code, disclosed training details or a permissive licence. Always check exactly what is available and what restrictions apply.
A developer downloads released model weights and runs them on their own computer.
Software whose source code is available for people to inspect, use and modify under a licence.
Open source describes the code and its legal permissions. In AI, a product may share code but not training data or model weights, so calling the whole system 'open' can be misleading unless the details are clear.
A developer can examine an open-source program, fix a bug and share the permitted changes.
The result produced by an AI system.
An output might be text, an image, a prediction, a tool request or a score. It should be checked according to the stakes of the task.
A generated meeting summary is an output.
When a model learns its training examples too closely and performs poorly on new cases.
An overfitted model has effectively memorised quirks instead of learning patterns that generalise. Testing on separate data helps reveal the problem.
A model scores perfectly on practice questions but fails when the wording changes.