AI scribes are amazing!

Started by Bartelby
Get help with your application

Use all the free resources available to you from SDN: articles, guides, expert advising, forums discussions, and school research.

I mean,
The big difference that LLMs often relies on neural networks, which is a form of machine learning that takes advantage of parallel processing (i.e., processing tokens more than one at a time) where "machine learning" can refer to something as simple as logistic regression. If you're feeding it examples and giving it feedback, you are training it to do a specific task, but you are not changing the method by which the model "learns."

I mean, you can go far beyond saying they 'often' rely on neutral networks, they really just are neural networks. And yes, I would agree with calling that training, although it is confusingly called 'post training' in the field. If it is not directly leading to modifications of the model however you are not causing it to learn anything that generalized beyond the particular instance of the LLM you are working with at that moment. Every time you create a new instance, you are at that point dealing with something that you haven't trained to do anything.

If they are genuinely doing post training at @Stagg737 's institution that would be impressive. However, they are then probably working with open weights models, because I promise you the frontier labs are not allowing third parties to do post training on their models. So technologically they are six months to a year behind.