As AI technology advances rapidly, the terminology can be confusing. Here's a quick guide to some of the key AI jargon I came across recently:
- Models - Large data sets that are trained on lots of data to understand prompts and questions. They vary in speed, price, size and quality.
- LLMs (Large Language Models) - Models specifically focused on understanding and generating natural language text.
- Providers - Services like OpenAI, Anthropic, Replicate, Cohere, Amazon Bedrock that provide easy APIs to test and run AI models.
- Hugging Face - A hub for open source machine learning models anyone can use. Has hundreds of models for text, image and speech tasks.
- Tokens - The way models count length of text prompts sent to them. More tokens used = more expensive.
- Embeddings - Turning text into mathematical representations models can match for similarity.
- Temperature - Setting that affects how creative/random a text generation model will be.
- Top percentile (Top p) - Similar setting that tweaks how deterministically the same a model's outputs will be.
- Fine tuning - Customizing a model by training it further on your own specialized data set.
- Prompts - What you send the model to prime it to generate a useful response.
- VectorDB - Database for searching embeddings to find similar documents.
- Streaming - Displaying a model's output as it generates, rather than waiting for it to finish.
- Evals - Test suites to evaluate model quality over time.
Top comments (0)