3 ms·

That's because the media hype is completely different from the industry hype. This article is a form of media hype. The pragmatic, already realized effects of

That's because the media hype is completely different from the industry hype. This article is a form of media hype. The pragmatic, already realized effects of machine learning are mostly in improving existing systems, especially classification systems, and have increased metrics by big percentages at big data companies, and more applications are being found every day. That is why real demand and interest is so high for ML. These generative, speculative models are interesting experiments by researchers but inconsistent and far overstudied by the media, because "AI invents new physics" is more understandable and sexier than "AI improves lead generation by 6.8%"

spamizbad 7y ago
I would say the industry definitely has its own hype. And honestly, I think the news media is a trailing indicator on that hype. For example, there's already an ongoing replication crisis that's been ongoing for several years with ML papers in computer science. I've spoken with people in the industry who have even speculated that there's straight-up academic fraud taking place. If you get a splashy, sexy finding you can get scooped up by a FAANG-like company with a fat salary before anyone gives your work a second look.
- omicron1 7y ago
  From working with many ML researchers, it's not speculation. Some aren't very proud, but if you want to get ahead and stay competitive, you have to publish your un-replicable results. The results aren't usually completely fake, but they depend on enough dishonest sleight of hand that they might as well be.
mobjack 7y ago
You can improve lead generation by 6.8% without ML too. A/B testing a few changes can easily accomplish the same thing. The AI could be optimizing to attract people who are willing to give out their email address than actual potential customers so it doesn't help the business. We tried using Google ads to optimize for leads before and we saw ads were on sketchy websites promising people free stuff. Lead generation was a bad KPI to use but it was chosen can't optimize for sales because the volume is to low to train the ML models. This is the reality of many businesses that experiment in AI and they produce junk. ML does work well for certain domains where you have the scale of data to pull it off, but now the hype has spread where it is not appropriate.

spamizbad 7y ago

I would say the industry definitely has its own hype. And honestly, I think the news media is a trailing indicator on that hype. For example, there's already an ongoing replication crisis that's been ongoing for several years with ML papers in computer science. I've spoken with people in the industry who have even speculated that there's straight-up academic fraud taking place. If you get a splashy, sexy finding you can get scooped up by a FAANG-like company with a fat salary before anyone gives your work a second look.

omicron1 7y ago
From working with many ML researchers, it's not speculation. Some aren't very proud, but if you want to get ahead and stay competitive, you have to publish your un-replicable results. The results aren't usually completely fake, but they depend on enough dishonest sleight of hand that they might as well be.

mobjack 7y ago

You can improve lead generation by 6.8% without ML too. A/B testing a few changes can easily accomplish the same thing. The AI could be optimizing to attract people who are willing to give out their email address than actual potential customers so it doesn't help the business. We tried using Google ads to optimize for leads before and we saw ads were on sketchy websites promising people free stuff. Lead generation was a bad KPI to use but it was chosen can't optimize for sales because the volume is to low to train the ML models. This is the reality of many businesses that experiment in AI and they produce junk. ML does work well for certain domains where you have the scale of data to pull it off, but now the hype has spread where it is not appropriate.