(2016) AI learns and recreates Nobel-winning physics experiment. [note: not a neural network!]

I've been thinking about how ludicrous it is that the doge Shiba Inu is the face of an agency that will be responsible for more deaths than Hitler, and I was wondering how TheFempire of old might feel about that.

Posted on TheFempire about a decade ago, here is the original link as archived at the time: https://web.archive.org/web/20160611145615/http://techcrunch.com/2016/05/16/ai-learns-and-recreates-nobel-winning-physics-experiment/

This was a few years after the neural network boom but before the 2022-era "AI == LLM == transformer model" so it's interesting to see how people talk about it. At the time, we used "AI" to refer to symbolic methods, to robotics, and broadly to machine learning approaches (although increasingly only neural-networks).

At the time, neural-network AI models were usually very exactly purpose-built and trained and inferred on just one GPU, sometimes two. What's notable is this is not a neural network model at all , the core algorithm is much more traditional machine-learning-and-optimization.

What they create is also very cool, literally: It's a "Bose Einstein Condensate", which you might remember if you learned "there are three or four states of matter" and wanted to find out if that was true.