I think the “Open Recent” menu debuted in 1980s? But now it is AI!
Everyone does it just to attach their old or boring tech to the fancy new trend and have some "cool" rub off on them.
I can think of a bunch of potential use cases for gpt 3 alone.
or do you mean its impossible to build useful models from scratch because all the "easy" problems are solved?
this also seems like a limited mind set.
context: I'm a ML noob
It's a large gap, covering everything from application topics, to data quality, to the need to actually run the damn think in a production setting with scalability, availability, error handling, etc.
Production applications of deep learning aren't particularly glamorous, they're not the "next big thing" right now. Rather they're improvements of existing applications.
Google's on device live captioning works really well, but still somewhat niche, and requires special / higher end SoC's to run.
models I have used seem to have their usefulness greatly outweighed by performance demands.
scaling and economics are another question entirely.
Perhaps we were spoiled with democratized web tech and it's wishful thinking to want everything to be that.
Search by image and object detection and computer vision in general is cool and potentially useful, but right now, it's cumbersome as fuck to pull out your phone, find the Lens application, take a picture etc. Needs to be baked into a wearable / neuralink type setup.
But self driving applications of CV work because the cameras are always deployed and running. But the hardware is expensive.
This is completely false. Here are some examples: Google translate. It's infinitely useful. It's not perfect, but it's good enough for me when I want to quickly check whether my translation into my second language is okay, or I'm not sure I got the meaning right of some translation.
Second example: My home security cameras now uses object classification and only alerts me when there's movement in "high risk zones" and it's human. So many stupid false positives of shadows and stray cats completely gone. I'm pretty sure I can fine-tune it with examples of myself and my family and it will ignore them when it's reasonably confident it's them, but I couldn't be bothered.
You work in a company making an intranet product for the paper tissue industry. Your manager wants v3.0 to have some AI in it. No one remembers the fiasco when "Cloud" was added in v2.0
you either swallow the CRUD app red pill, or you live long enough to become the "AI a la carte" manager
A fun game used to be "translate this phrase English to French, then translate it back again and laugh at how meaningless it's made by the round-trip". That doesn't work any more.
However it was all done with hand-crafted statistical functions and code. The new stuff using deep learning is the first time it'd have been referred to as "AI".
However, what I think the poster meant was that just tacking on AI for the sake of it to a problem which most likely doesn't have an applicable use for it (which currently is the case for most existing IT projects or apps) is doomed for failure.
Do this:
1. make something that has some "modern" AI such as deep learning in it, but that doesn't do anything much useful (because that's almost impossible in most cases). E.g. predict something trivial based on a time series, and repoort it. We all know how hard this is, so just take some hello world ML project and change some inputs.
2. Find some easy project that actually makes business sense, and that isn't yet done, and can be done with simple logic or some old school "traditional" ML or constraint solving e.g. a scheduling/optimization problem.
3. Make sure these efforts are "merged" so that the fancy AI tech AND the measurable good outcome is the same project. The "AI" moonshot of the company.
4. You can now say you are doing "AI", you have measurable success.