# Training

Concept · [Building with AI](https://ecosystem.samcarlton.com/ai)

Also called: model training, trained, training data, pre-training, pretraining

Teaching an [AI model](https://ecosystem.samcarlton.com/model) by showing it a very large number of examples before anyone uses it. The first, largest round is called pre-training. [Fine-tuning](https://ecosystem.samcarlton.com/fine-tuning) is extra training on a smaller, narrower set of examples.

**For example:** An [LLM](https://ecosystem.samcarlton.com/llm) is trained on a vast amount of text once. Later, a company might [fine-tune](https://ecosystem.samcarlton.com/fine-tuning) it on its own support [conversations](https://ecosystem.samcarlton.com/session) so it answers in the company's style.

## Connects to (27)

- [LLM](https://ecosystem.samcarlton.com/llm): Large language model.
- [Inference](https://ecosystem.samcarlton.com/inference): An AI model producing an answer to a request.
- [RAG](https://ecosystem.samcarlton.com/rag): Retrieval-augmented generation.
- [Gradient descent](https://ecosystem.samcarlton.com/gradient-descent): The method used to train neural networks: measure how wrong the model is, work out which direction to nudge each weight to be less wrong…
- [Loss function](https://ecosystem.samcarlton.com/loss-function): A formula that scores how wrong a model's output is compared with the right answer.
- [Neural network](https://ecosystem.samcarlton.com/neural-network): A kind of software model made of many layers of simple math units connected by weights, loosely inspired by the brain.
- [RLHF](https://ecosystem.samcarlton.com/rlhf): Reinforcement learning from human feedback.
- [NVIDIA](https://ecosystem.samcarlton.com/nvidia): NVIDIA makes GPUs, chips that help computers train AI models and run them to produce answers.
- [Model](https://ecosystem.samcarlton.com/model): The trained system at the center of an AI product.
- [Fine-tuning](https://ecosystem.samcarlton.com/fine-tuning): Extra training of an existing model on a smaller, narrower set of examples, so it suits one task or style.
- [Parameters](https://ecosystem.samcarlton.com/parameters): The values a model learns during training.
- [GPU](https://ecosystem.samcarlton.com/gpu): Graphics processing unit.
- [Knowledge cutoff](https://ecosystem.samcarlton.com/knowledge-cutoff): The date where a model's training data ends.
- [Next-token prediction](https://ecosystem.samcarlton.com/next-token-prediction): The one job a language model performs.
- [Transformer](https://ecosystem.samcarlton.com/transformer): The design that most modern language models are built on, introduced by Google researchers in 2017.
- [Parametric knowledge](https://ecosystem.samcarlton.com/parametric-knowledge): What a model knows from its training, stored in its parameters.
- [Sycophancy](https://ecosystem.samcarlton.com/sycophancy): The habit AI models have of telling you what you want to hear: agreeing too readily, praising weak ideas or giving up a correct answer…
- [MLOps](https://ecosystem.samcarlton.com/mlops): The practices and tools for running machine learning models in production: versioning data and models, deploying them, monitoring quality…
- [Deep learning](https://ecosystem.samcarlton.com/deep-learning): Machine learning using neural networks with many layers.
- [Overfitting](https://ecosystem.samcarlton.com/overfitting): When a model learns its training examples too closely, including their quirks, and then performs poorly on new data.
- [Distillation](https://ecosystem.samcarlton.com/distillation): Training a smaller model to copy the outputs of a larger one, so it keeps much of the quality while being cheaper and faster to run.
- [MicroGPT](https://ecosystem.samcarlton.com/microgpt): A very small GPT language model written by Andrej Karpathy in a few hundred lines of plain Python, meant to show how training and running…
- [Python](https://ecosystem.samcarlton.com/python): A programming language known for being easy to read.
- [CUDA](https://ecosystem.samcarlton.com/cuda): NVIDIA's platform for writing software that runs on its GPUs.
- [CS50 AI](https://ecosystem.samcarlton.com/cs50-ai): Harvard's free online course on the ideas behind modern AI, taught in Python, with topics such as search, machine learning and large…
- [NVIDIA Inception](https://ecosystem.samcarlton.com/nvidia-inception): NVIDIA's free program for AI startups.
- [AI addendum](https://ecosystem.samcarlton.com/ai-addendum): An add-on to a software contract for a product that uses AI.

Map view: https://ecosystem.samcarlton.com/training