Not OP, but i took that statement to mean it's the observed quest for humanity. In every step we put great effort into ensuring we aren't the ones who work. Be it slaves, harnessing electricity, general technology and etc.
We do seem to strive to push towards extremes of optimization. Consequences be damned.
That's a reasonable observation, but there's a big distinction between "aren't the ones who work" and "obsolete".
My experience makes me think that people very much do _not_ want to be obsolete, and that being obsolete is devastating to one's sense of identity and social acceptance. I do think they also have the drive to work less, but there are plenty of people who do things for fun that other people would call "work".
> Not OP, but i took that statement to mean it's the observed quest for humanity. In every step we put great effort into ensuring we aren't the ones who work. Be it slaves, harnessing electricity, general technology and etc.
This is true. Humans have always strived to make their own lives easier. It is just that the rewards of an easier life are not being distributed to the actual people doing the work. Instead, a disproportionate amount of the gains have been going to the investor class.
Things that happen are rarely result of a decision. And even when it seems like decision was made in actuality it was just a result of historical, geographical, economical and technological factors. If this person didn't make that this decisions then, someone other would promptly.
> Mr Liu also said Foxconn is trying to "convert itself from a manufacturing service company to a platform solution company," citing smart cities and smart manufacturing as other applications for AI factories.
Its seems like China is trying to speed run modern capitalism they are now moving from the manufacturing stage to the "spew constant BS in order to deceive investors stage".
Taiwan is basically as-if the Confederacy fled from the Union and took over Cuba. And then both the Confederacy and Union laid claim to the United States except obviously the Union actually has physical control over it. Also imagine that the rest of the world was pro-slavery so they liked the Confederacy more so initially all the countries only recognize Taiwan as the actual "United States" but its obviously more practical to recognize the Union (China) as the actual "United States" as you start doing a ton of business.
Basically replace Union with CCCP, Cuba with Taiwan, and pro-slavery with Democracy (CCCP being Communists).
More like if the union lost and fled to Cuba. But there is really no point in trying to making this comparison. They are only similar in that both are civil wars. Beyond that they are different in almost every aspect: from the initial conditions to the war itself to its end.
I think the TL;DR is foxconn wants to assemble data center/a bunch of servers with Nvidia's chips which they call AI factories, but the marketing BS is strong.
How future proof is all this stuff against changes in AI? What if someone discovers a viable alternative to ANN's or a way to implement them better that doesn't benefit from GPU's?
That's one of the benefits of GPGPUs vs custom training silicon. As long as the solution requires highly parallelizable calculations a GPU will be able to accelerate it.
On the contrary, dendrites/synapses/axons implement something that looks an awful lot like matrix multiplication followed by an activation function. The biggest differences between bio and artificial neural networks -- spiking, backpropagation, etc -- seem to be in the non-matmul parts of the story.
But it is not matrix multiplication. We want it to be matrix multiplication, we want to use a minimalistic, elegant, reductionist approach, we want to claim we understand it, but in practice it's none of that otherwise we would have a working artificial brain.
As we discover things we find those things at work in the brain, but the brain is too advanced for us to make proper sense of it.
We only understand a small part of a big story, but we do understand that small part and it does look like matrix multiplication. Most of what goes on in the brain is up for debate. But not that part.
Incidentally, this is the whole reason why reductionist approaches are so important: without reductionism, you can never get a foothold. You are adrift in a sea of woo and can never make any progress towards understanding because one nonfalsifiable holistic theory is as good as the next. However, if you break the system down into pieces until you do understand a piece, you can start to put the puzzle together. Dendrites, synapses, action potentials, thresholds, potentiation, plasticity -- these things are real, we do understand them under an ever growing set of circumstances, and while the picture on the puzzle is still anyone's guess the corner that looks like matrix multiplication is fully assembled and not going anywhere.
In the current world, deep learning with homogeneous computation graphs, tuned with backprop, has won the Hardware Lottery [1]. This is unfortunate for research outside of that area, but just looking at the momentum of development it seems like a sure bet to keep investing in GPU-based training and inference for the next decade. There's just too much lock-in already to this paradigm.
If a new algorithm appears from with a novel approach (analog compute, heterogeneous computation graphs from genetic algorithms, quantum, much more...), there will be a whole generation of R&D + tool + framework building, which gives the major players enough time to adapt.
Graduated too late to explore the blockchain.
Graduated too early to experience fully automated luxury space communism.
Graduated just in time to work in an AI factory
39 comments
[ 586 ms ] story [ 2205 ms ] threadWe do seem to strive to push towards extremes of optimization. Consequences be damned.
My experience makes me think that people very much do _not_ want to be obsolete, and that being obsolete is devastating to one's sense of identity and social acceptance. I do think they also have the drive to work less, but there are plenty of people who do things for fun that other people would call "work".
This is true. Humans have always strived to make their own lives easier. It is just that the rewards of an easier life are not being distributed to the actual people doing the work. Instead, a disproportionate amount of the gains have been going to the investor class.
A friend was recently explaining to me, that they seriously consider the impact of things like tech, automation, etc on jobs.
So, often they decide to not automate stuff in order to keep the jobs for their workers. Sounds like better workplace stability.
Its seems like China is trying to speed run modern capitalism they are now moving from the manufacturing stage to the "spew constant BS in order to deceive investors stage".
Basically replace Union with CCCP, Cuba with Taiwan, and pro-slavery with Democracy (CCCP being Communists).
Quick maths.
As we discover things we find those things at work in the brain, but the brain is too advanced for us to make proper sense of it.
Incidentally, this is the whole reason why reductionist approaches are so important: without reductionism, you can never get a foothold. You are adrift in a sea of woo and can never make any progress towards understanding because one nonfalsifiable holistic theory is as good as the next. However, if you break the system down into pieces until you do understand a piece, you can start to put the puzzle together. Dendrites, synapses, action potentials, thresholds, potentiation, plasticity -- these things are real, we do understand them under an ever growing set of circumstances, and while the picture on the puzzle is still anyone's guess the corner that looks like matrix multiplication is fully assembled and not going anywhere.
If a new algorithm appears from with a novel approach (analog compute, heterogeneous computation graphs from genetic algorithms, quantum, much more...), there will be a whole generation of R&D + tool + framework building, which gives the major players enough time to adapt.
[1] https://arxiv.org/abs/2009.06489