DeepMind's Losses and the Future of Artificial Intelligence

[GRULSTMRNN]

You are absolutely right, in 40 years people will use way more advanced stuff than today and today use way more efficient and advanced techniques than 40 years ago. The stuff from 40 years ago does not matter today. The only difference is that I won't be running around like a maniac in 40 years and tell everyone what a hot shit I was back then. And I as sure as hell won't make myself to a laughing stock by telling people that the stuff they use in 40 years is basic when most likely my entire annual medical checkup will be conducted by an algorithm, surgeries by algorithms, autos driven by an algorithm. Equally today the stuff that enables autonomous cars is amazing and so are algorithms that already apply the currently most advance game theory concepts. The stuff 40 years ago is crap in comparison.

It's called technological advance mate. End of story.

We used assembly, C, and BASIC. You can write anything, including basic neural network in any programming language. I didn’t say they were advanced or deep or had billions of data pints, or anything else at that time. AI was advancing since then, and it encompasses machine learning, so when ML and NNs were hot then people talked about AI. 40 years from now people may not remember what was DNN or RL, and may use a different form of AI, arguing that 40 years ago AI couldn’t have been hot. Quantum computers are also hot for a while, every 5-10 years.
And I’m very aware of AI capabilities today, and I’m not saying that it’s useless, or that it will be useless in the future. It’s useful now but it has its limits, which right now sometimes can be solved without AI, or accepted as limitations. I do have plans to use AI/RL myself but can’t know yet whether it will do better than I can do without it. Unlike you, I’m not a psychic who can predict future, you hot shit.

 

[guru]

You are absolutely right, in 40 years people will use way more advanced stuff than today and today use way more efficient and advanced techniques than 40 years ago. The stuff from 40 years ago does not matter today. The only difference is that I won't be running around like a maniac in 40 years and tell everyone what a hot shit I was back then. And I as sure as hell won't make myself to a laughing stock by telling people that the stuff they use in 40 years is basic when most likely my entire annual medical checkup will be conducted by an algorithm, surgeries by algorithms, autos driven by an algorithm. Equally today the stuff that enables autonomous cars is amazing and so are algorithms that already apply the currently most advance game theory concepts. The stuff 40 years ago is crap in comparison.

It's called technological advance mate. End of story.

I’ll just tell you a small secret then: I wasn’t hot shit back then. I was just a typical student doing my homework assignment.
Though I don’t mind that you like to make assumptions and then laugh at them. In the past you also tried to make assumptions about my nationality. And even here you assumed that I disagree with you because I posted an article that you didn’t like, while you don’t actually know whether we agree or disagree, and about what.
That’s all for today, now gotta work on some hot shit.

 

[tommcginnis]

ML != Deep Learning.

Errrrr -- Nowhere did I make such a statement. You have drawn that inference all on your own -- from the same pissy black well from which you've sought to attack guru. Sad, that. About a quarter of your posts show real insight and knowledge-born-of-experience. Which leaves 3/4s being vacuous rantings.

["Hmmmm. Maybe his training data are unstable?" Indeed, my little friend. Indeed.]

 

[TommyR]

think this was more joke and for the pr. in any case this would be considered an excellent return on investment. they've got people with beards who design tpu's for serious stuff.

 

[GRULSTMRNN]

Ready to return to facts? Which part that I stated do you factually disagree with? Or is it just that you get extremely uncomfortable each time when there is disagreement in the room?

I talked about DNN then you come out of nowhere and start talking about ML in the 80s. So my DNN!=ml was a direct reference to you replying to my specific post on a completely different topic. I never talked about ml and never denied that it was applied decades ago. Then you go on with with some scifi movie. Perhaps you were just drunk? "No more ET for this man after 2 beers"

Errrrr -- Nowhere did I make such a statement. You have drawn that inference all on your own -- from the same pissy black well from which you've sought to attack guru. Sad, that. About a quarter of your posts show real insight and knowledge-born-of-experience. Which leaves 3/4s being vacuous rantings.

["Hmmmm. Maybe his training data are unstable?" Indeed, my little friend. Indeed.]

 

[Simples]

Ready to return to facts? Which part that I stated do you factually disagree with? Or is it just that you get extremely uncomfortable each time when there is disagreement in the room?

I talked about DNN then you come out of nowhere and start talking about ML in the 80s. So my DNN!=ml was a direct reference to you replying to my specific post on a completely different topic. I never talked about ml and never denied that it was applied decades ago. Then you go on with with some scifi movie. Perhaps you were just drunk? "No more ET for this man after 2 beers"

View attachment 207719

https://en.wikipedia.org/wiki/Deep_learning

For you it might be useful to make distinctions, though common understanding of ML is any algo capable of learning (from unsupervised to supervised). With specialized hardware, evolving architecture/designs, of course, experts will want to make differentiations within their field. Common understanding would also say it's all AI, and they're partially right, and also completely wrong. One wouldn't call a few regressions AI, and also, AI has always been a moving target.

 

[GRULSTMRNN]

Very fair point and I can fully agree. I just found it funny when some dinosaurs belittle the incredible advances that were made in deep learning over the past few years. Advances that were simply not possible before because of limited hardware and limited training data. When the world's leading philosophers, economists, and scientists break their heads over the question how a majority of humans will remain competitive and make a living then it is preposterous when some anonymous lemmings think they know better and there is nothing to fear and that ai will not solve some of the pressing problems of our times. Of course they will come out and twist my words again suggesting I said that ai will solve all problems (which I did not say) and that they did not declare ai of little value (something they exactly meant to say).

https://en.wikipedia.org/wiki/Deep_learning

For you it might be useful to make distinctions, though common understanding of ML is any algo capable of learning (from unsupervised to supervised). With specialized hardware, evolving architecture/designs, of course, experts will want to make differentiations within their field. Common understanding would also say it's all AI, and they're partially right, and also completely wrong. One wouldn't call a few regressions AI, and also, AI has always been a moving target.