Think Together · Start with a Short Story
The girl wants to know where chat skills come from. The robot draws the process as “first learn many patterns, then learn to do a task better.”
Lay Out the Process
- Prepare dataClean and organize it
- PretrainLearn broad patterns
- Later trainingImprove task and conversation behavior
- EvaluateCheck abilities and problems
Language model training usually includes learning broad patterns in data and later training for target behaviors.
A training goal may ask the model to predict what comes next or fill in missing content, gradually adjusting its parameters for those tasks. Examples and preference feedback can later improve how it responds.
Exact steps vary by model. Data quality, computing resources, and evaluation all matter.
Now You Try
If a model finished training yesterday, is it guaranteed to know today's news?
Ready? Open this to see the explanation
No. To learn about a new event, it needs new reliable information, such as an appropriate search, and the sources should be checked.
One Idea to Take Away
Language model training usually includes learning broad patterns in data and later training for target behaviors.
Try covering the diagram above and explaining each step to someone in your family. If one step is hard to explain, return to that card and look again.