Ask HN: Are we in an AI / ML bubble?
As in title. Is there a systemic over-valuation of companies in these spaces?
I took AI/DL course before and professor demonstrated from scratch how to build an algo and train model capable of amazing upsampling of imagery.
Hence - bubble or not - there are lots of useful, practical benefits to this today.
There is likely a mis-prioritization of applications in the market.
You should look at bubbles in the context of applications, and perhaps dependent on strategy separately. Context ~ Health, AutoBI, NLP...; Strategy ~ unsupervised, explainable, deep, symbolic...
Another way to look at this is identifying Misery stocks. If the economy takes a hard dip, what will continue to be funded. Those are misery stocks. Some AI will survive misery at varying levels, others will not.
Other companies that brand themselves as AI do so for PR purposes. Most of them don't use AI, use robust techniques that have been around for many decades, or use it only to justify their pitch. Their product would be just the same with rule-based systems or without any kind of ML.
I would add, there are some DL/ML business components which are in heavy use in some industries which given the companies size are giant productions deployments of "IA" (not expert systems or other automation technologies), so you could say we are actually living somehow in the principle of a golden age of "IA" (DL/ML techniques), but DL/ML techniques - at least what is known publicly to be the state of the art - has some practical limits (i.e. power consumption to traing useful models), but workarounds for those limits are being heavyly studied or solutions are being tested (as we speak indeed).
What's here for sure, it's a golden age of data: you can extract (meta)data from almost everything running on a CPU/GPU, the "likes" in everything are the users training models, not exactly the models you imagine (because you can associate/correlate some scenarios - "a birthday party at the office" - with others you would think are a lot different - "a christmas party at the bar" - but they'are not so different actually, and the features found by the training are more or less - it should be a % - interchangeable.
So yeah, every "like" out there is training something behind the scene.
If this hype-cycle doesn't sound ridiculous, I've also heard that my local grocery store is using deep-learning + kubernetes to scale their supply chain systems. They told me that they're writing a Medium blog-post about this which is coming soon /s
Few companies build products that generate or interface with data at that level.