Daniel LowdRemember when most ML applications were some variation on surveillance and advertising?
Steve Leach@lowd I remember when most ML applications were variations on #MNIST. And #Imagenet, but I only had enough computer at the time to play around with Mnist. But yea, even then "Recommendation Engines" were starting to be the first things anyone mentioned because it was low hanging fruit - something of immediately obvious commercial value with terrific training data and an easy task for deployment.
@stevenaleach I don’t consider MNIST to be an application — more of a benchmark for computer vision. In terms of what ML engineers actually did in industry, it seems like a high percentage was ads and surveillance. Recommender systems have a lot of value, but are at least advertising-adjacent in practice.
(Caveat: my view on what actually happens in industry is limited. This is just my perception.)
@lowd Exactly, in 2015/2016, the AI spring had already been sprung but everything was still focused on those benchmarks. This was around the time GANs were the new great thing, the beginning of generative models - but there wasn't commercialization yet. But the one thing that *was* discussed frequently already was recommendation engines - because it was the obvious low hanging fruit that could be commercialized first.
@lowd And no, I have no industry experience either. I'm a hobbyist basically. Currently I'm a recently turned ex-dishwasher now a part-time pizza-dough maker '-). But I first got into machine learning around 2015 when #Theano was still the leading platform and I'm pretty sure #Tensorflow didn't exist yet. It's been fun watching the field grow from where it was then, but also frustrating because marketing is the "killer-app" so far as the field has turned into an industry.