AI technologies for the web:

Options to run AI models:

API call to LLM provider
- Examples:
  - OpenAI's API: ChatGPT
  - Google's API: Gemini
  - Anthropic's API: Claude
- Benefits: Easy
- Drawbacks: Cost
Host own LLM on server
- Examples:
  - https://github.com/bentoml/OpenLLM
    - Llama 2 (Meta) --> Free
    - Mistral (Mistral AI) --> Paid?
- Benefits: Security, Privacy, Customisation, Avoid vendor-lock-in
- Drawbacks: Setup costs, Updates
LLM on client
- Examples:
  - Transformers.js: https://huggingface.co/docs/transformers.js/index
    - https://syntax.fm/show/740/local-ai-models-in-javascript-machine-learning-deep-dive-with-xenova
  - TensorflowJS: https://www.tensorflow.org/js
    - https://www.tensorflow.org/js/models
  - LLM Inference API: https://ai.google.dev/edge/mediapipe/solutions/genai/llm_inference/web_js#create_the_task
    - https://developer.chrome.com/blog/io24-web-ai-wrapup
  - Benefits: Costs (free models)
  - Drawbacks: Download time
LLM in Browser (https://developer.chrome.com/docs/ai/built-in): This might be the next area where browsers try to get ahead. Google integrates Gemini Nano. Will Apple work together with OpenAI. Apple has devices and OpenAI the techonlogy.

Install library: https://huggingface.co/docs/transformers.js/main/en/installation
Use pipeline API and optionally provide model. Models can be implicit. Explicit models can be found by clicking on one of the tasks.
See more examples: https://huggingface.co/docs/transformers.js/main/en/index#examples