Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.

2 Comments

  1. Maybe this is possible in a different context: not during prontos, but using supervised learning during training.

    I believe chat GPT uses unsupervised learning algorithms: feed it lots of text, have it generate an input, and somehow check of that output is good (usually using other AI).

    But classical neural networks use supervised learning: “given this input x, you should output y”.

    Maybe we can have an hybrid approach: unsupervised first with millions of texts, supervised later with a few hundred prompts and expected answers, crafted by a writer to ensure some quality.