In computer chess because:
1. the mechanism they are using, "Transformer" machine learning models (
link), are exceptionally versatile at learning, and can easily be used in a wide variety of learning tasks (e.g. they have been trained to play chess -
link). The current version of TensorFlow has this type of learning available.
2. the science forum is still closed
A new startup, including several former Google employees, has a new, and somewhat different, approach to getting us closer to machine general intelligence. They are going to be using Transformer machine learning models - a relatively recent development in neural networks. They are useful for their versality.
The first task they will be working on is... operating software on your computer for you! A kind of Google Assistant with superpowers! It seems like a good idea to me. I wonder how long before you can say,
"Build me a chess program that plays like CS-Tal", and have it ready the next day? That will be some time - they've already said they're going to start simple and work from there.
Looks like an interesting company to keep an eye on!
https://www.theregister.com/2022/04/27/adept_ai_google/
Home 