Why this matters
Two terms — inference and machine learning (training) — make it much clearer how CaraOne turns raw footage into searchable intelligence. Here is the distinction, followed by how CaraOne uses each.
What "inference" means
Inference is the process of drawing conclusions or making predictions from data. Its exact meaning depends on the kind of AI.
Traditional (rule-based) AI
Rules and logic are explicitly programmed by people. Inference is the logical deduction of new conclusions from those handcrafted rules. These systems are deterministic and do not learn from data.
Machine learning
A model first learns patterns from training data. Inference is then applying that trained model to new, unseen data to make a prediction — without learning anything further in that moment.
- Supervised learning — trained on labeled input/output pairs; inference predicts outputs for new inputs.
- Unsupervised learning — trained on unlabeled data to discover patterns; inference clusters or simplifies new data.
- Reinforcement learning — an agent learns from rewards and penalties; inference uses the learned policy to act in new situations.
The key differences
| Traditional AI | Machine learning | |
|---|---|---|
| Knowledge | Explicit, human-written rules | Patterns learned from data |
| Flexibility | Rigid; rules edited by hand | Generalizes to unseen examples |
| Learning phase | None; rules are predetermined | Models adjust parameters from data |
How CaraOne uses this
CaraOne is a machine-learning system, and both phases run entirely on your own hardware — nothing is sent to the internet.
- Inference = indexing & search. When CaraOne watches your media, it runs its trained models to extract meaning (visuals, spoken word, faces) and to answer searches, predictions, and Ask CaraOne queries. No learning happens during this step.
- Training / finetuning = teaching CaraOne something new. CaraOne ships with broad world knowledge, but it can be fine-tuned on your own data to recognize domain-specific concepts or objects. That is a separate, deliberate step — Ask us to see a copy of Training CaraOne for recognizing new concepts.
In short: day to day, CaraOne performs inference on your media; when you need it to learn something new, that is training.