Hacker News Reader

Comment by hn_throwaway_99 | Hacker News Reader

So I guess "A.I." is the new buzzword for how we describe all digital technology going forward. FWIW I thought this was a better article that described the actual tech used by KoBold, https://spectrum.ieee.org/ai-mining. Now, that article was written by the KoBold CEO, so there are certainly parts of it I'd take with a giant chunk of salt, but I think it's easier to read that article (and read past some of the AI buzzwords) to see how they're probably using (a) better surveying tech and (b) standard machine learning techniques to generate maps of potential deposits.

View on HN

I generally dislike the term AI because it could reasonably describe most computer programs.

However, in this case machine learning is involved, so even by a narrower definition calling it "AI" seems more than fair.

> So I guess "A.I." is the new buzzword for how we describe all digital technology going forward

I wonder if any companies are getting deals on compute for making a big splashy deal out of the part ML played in these processes. Kind of a B2B meets Twikstogrube Influencer marketing strategy, but instead of companies having cachet because a bunch of social media followers find them appealing, they actually manifest things in the physical world. That is a big hole in the Generative AI company sales pitch for a) non-early-adopter potential customers, and b) many others looking uneasily at the kind of resources they're tearing through when the only tangible things they've seen from it are a pitches for features they never asked for and don't care about, and very concerning faked images and videos for extortion, bullying, porn, and political shit. I'm not saying those things are all its good for, but the communication about the real-world value of this stuff has been pretty lacking, and the drawbacks have been understandably shouted from the rooftops, so they're probably preeeetttyyy thirsty for stories like this.

Ml became ai because it's the base for ai.

It's just what it is.

Nonetheless LLM Made it a lot easier for people to understand that investment in ml is really really helpful

“That means the conventional predictions are largely inference—and worse, they result in unquantified uncertainty.”

Wild claim given the fact that Gaussian process regression / Kriging was invented in the 1960s in geoscience to do exactly what the article claims only their models do: “quantify uncertainty, which in turn guides our data collection, as the most uncertain rocks often represent the most valuable ones to sample”

IMO AI means neural net. I get that people use it for other things, but that's what I use it to mean - there is just no other term that's easy to say. And at this point the idea of breaking problems down into "neurons" and activation patterns is inherent to most AI models. Here though the keywords are "ensemble machine learning" and "Bayesian" - they could have used a neural net for the machine learning but most likely it is just XGBoost or similar. https://ia.acs.org.au/article/2021/the-ml-technology-looking... mentions they are also doing full-physics joint inversions and computer vision, perhaps the vision is a neural net.

Historically however AI didn't mean a neural net specifically.

Yeah, historically AI incorporates most of modern computer science, dating back to the 1950s.

No, not really. If you compare proceedings of even the earliest AI and CS conferences it is clearly not the case.

Can you expand on that with some examples or links?

I dug into this a tiny but and learned that the ACM started having conferences in the early 1950s - https://dl.acm.org/doi/pdf/10.1145/30408.30418 - while the famous Dartmouth workshop that coined the term AI was later, in 1956. So I agree, it looks like computer science was already establishing itself as a discipline before the term AI started to be used.

There's been some overlap naturally, e.g. Yershov or Hopper attending both types events early on. But hardly any of content in say TAoCP was considered AI at any point.

my friend, XGBoost is not a neural net-based method, it represents a best-of-breed for an alternative family of methods.

"While the XGBoost model often achieves higher accuracy than a single decision tree, it sacrifices the intrinsic interpretability of decision trees. For example, following the path that a decision tree takes to make its decision is trivial and self-explained, but following the paths of hundreds or thousands of trees is much harder."

https://en.wikipedia.org/wiki/XGBoost

Unless the parent edited their comment, you misread it, because they are specifically saying XGBoost is not a neural network:

> Here though the keywords are "ensemble machine learning" and "Bayesian" - they could have used a neural net for the machine learning but most likely it is just XGBoost or similar.

I.e. they could have used a neural network but they probably used something else, that something else being XGBoost.

The article you linked was written by the CEO of the company.

So it’s not going to go into detail about possible negative impacts of the mining in the same way that the original NYtimes article does.

Reply on news.ycombinator.com