3 ms·

AI is: - It's expensive / difficult to operate / only used by those in the know - It doesn't do half as much as you can hope it will do. - Eventually it will

by d--b 4y ago

AI is: - It's expensive / difficult to operate / only used by those in the know - It doesn't do half as much as you can hope it will do. - Eventually it will be ubiquitous.

supermatt 4y ago
AI is neither expensive nor difficult to operate. You likely unknowingly use AI multiple times a day - so it clearly can't be difficult OR expensive if you don't even notice that! Im assuming you mean the many GPU hours that go into training the HUGE ML models like DALL-E, etc. That is FAR from the norm. After following online courses I was able to solve real problems in my domain of interest with nothing more than a few hours training with a relatively small (<500MB) dataset on my PCs GPU. It cost WAY less than a dollar to train, and is now used by thousands of unskilled users on their mobile devices daily.
bluejay2387 4y ago
I frequently prototype models on my Android tablet when on flights. You can install Jupyter on almost anything and even low end notebooks have the processing power to train useful (albeit small) models. Google Collab and PaperSpace have free tiers and for only $10/month you can get access to GPU's on PaperSpace with the capacity to train fairly decently sized models. There is a wealth of intro material (this one is pretty good https://www.oreilly.com/library/view/hands-on-machine-learning/9781492032632/ https://www.oreilly.com/library/view/hands-on-machine-learni...) to get you started. Even with the largest models that are not feasible for individuals to train (Large Language Models for example, that are so expensive even most companies can't afford to build them) there are refinement techniques that let you take existing models and train them for specific tasks. That kind of training can run on most higher end consumer grade GPU's. I'll admit I am bit biased since I have worked in the field of AI for 20 years now, but I really don't see anything limiting even a mildly committed individual from getting into this area, even if only as a hobby.
- d--b 4y ago
  > I really don't see anything limiting even a mildly committed individual from getting into this area, even if only as a hobby. My point exactly: it shares a lot of similarity to PCs in the early 90s

by d--b 4y ago

AI is: - It's expensive / difficult to operate / only used by those in the know - It doesn't do half as much as you can hope it will do. - Eventually it will be ubiquitous.

supermatt 4y ago

AI is neither expensive nor difficult to operate. You likely unknowingly use AI multiple times a day - so it clearly can't be difficult OR expensive if you don't even notice that! Im assuming you mean the many GPU hours that go into training the HUGE ML models like DALL-E, etc. That is FAR from the norm. After following online courses I was able to solve real problems in my domain of interest with nothing more than a few hours training with a relatively small (<500MB) dataset on my PCs GPU. It cost WAY less than a dollar to train, and is now used by thousands of unskilled users on their mobile devices daily.

bluejay2387 4y ago

I frequently prototype models on my Android tablet when on flights. You can install Jupyter on almost anything and even low end notebooks have the processing power to train useful (albeit small) models. Google Collab and PaperSpace have free tiers and for only $10/month you can get access to GPU's on PaperSpace with the capacity to train fairly decently sized models. There is a wealth of intro material (this one is pretty good https://www.oreilly.com/library/view/hands-on-machine-learning/9781492032632/ https://www.oreilly.com/library/view/hands-on-machine-learni...) to get you started. Even with the largest models that are not feasible for individuals to train (Large Language Models for example, that are so expensive even most companies can't afford to build them) there are refinement techniques that let you take existing models and train them for specific tasks. That kind of training can run on most higher end consumer grade GPU's. I'll admit I am bit biased since I have worked in the field of AI for 20 years now, but I really don't see anything limiting even a mildly committed individual from getting into this area, even if only as a hobby.

d--b 4y ago
> I really don't see anything limiting even a mildly committed individual from getting into this area, even if only as a hobby. My point exactly: it shares a lot of similarity to PCs in the early 90s