It has been really interesting to see how GPT-3 is bringing a new paradigm to Machine Learning. It is an incredibly compact representation on essentially all data in the internet.

#GPT3 #MachineLearning #AI

1/5

But most interestingly, it operates in a different way. People aren't creating those amazing demos by fine-tuning the demos with their own data. No.

They are providing a few "priming" examples that "reconfigure" the model to that task right then and there.

2/5

This is a great pattern matcher over all the internet's data so it essentially makes the model target the corresponding space of the inputs you provided it.

3/5

This can bring a real paradigm shift to Machine Learning and, in particular, in how transformer models can be used by businesses.

4/5

Moving to a Priming mentality rather than a Training mentality could open some new doors to businesses and hackers to use Machine Learning in completely new, innovative ways.

This is a great thread summarizing it as well: https://twitter.com/sh_reya/status/1284545976892403714

Thoughts?

5/5