Hacker News Reader

Comment by brndnmtthws | Hacker News Reader

They've got a lot more people, so it's just a matter of time. That, and there aren't any real "trade secrets" when it comes to AI. It's all just computer code, and anyone with an Internet connection can teach themselves how to use Wikipedia and run TensorFlow.

View on HN

'It's all just computer code'

Are you saying no software has trade secrets?

Sure, you can go through some tutorials and learn to build a CNN, or even read more recent research papers and hit bleeding-edge ImageNet benchmarks, but nothing currently public is going to get you close to Google's SDC.

You can't pretend this is just about people either, India's at 1.3B and they're far from becoming a world leader.

This really depends on what is the driver of the current rise of neural networks and other similar machine learning techniques... whats the contributions of data and method toward the new improvements in performance? If more data is main contribution I would venture to say that China can indeed beat USA.

It's a combination of recent innovations and more data. Having access to more annotated/labeled data has definitely propelled the state of the art in some disciplines like machine vision.

Most of the time, secrets and other forms of protectionism are techniques used for covering up incompetence, or hiding bad behaviour.

bold claim. citation needed for "most of the time".

common counterpoints: secrets can be operationally necessary, secrets can create economic value.

Wait, what does Wikipedia have to do with TensorFlow? I think you are over-simplifying AI. Yes, you can take pre-trained models are do useful things with them. But creating these models and training them initially on vast quantities of data still remains a challenge and has to be overcome by real engineers.

Not anyone can do AI at scale.

> Not anyone can do AI at scale.

For that you'd have to wait until next year. Many low level skills necessary in ML are being automated. It's getting closer to being easily used by hackers.

I'd like to see someone hack together AlphaGo.

TensorFlow is the trendiest ML library, and Wikipedia is a source of knowledge. You can gain the knowledge required to effectively use TensorFlow (or whatever) from freely available sources, provided you have the interest and tenacity.

More people applies to many applications. Drug research, for example, would greatly increase if China got into that business.

Google's PageRank code is a trade secret.

Reply on news.ycombinator.com