3 ms·

Every financial bubble has moments where, looking back, one thinks: How did any sentient person miss the signs? Well maybe a lot of people agree already with w

Every financial bubble has moments where, looking back, one thinks: How did any sentient person miss the signs? Well maybe a lot of people agree already with what the author is saying : the economics might crash, but the technology is here to stay. So we don't care about the bubble

rwmj 1y ago
The technology is there til the GPUs become obsolete, so about 3 years.
- ttoinou 1y ago
  Our current GPUs can last a decade or more, no ?
  - rwmj 1y ago
    They (probably) won't physically fail, but they'll be obsolete compared to newer GPUs which will have more raw compute and lower power requirements.
_heimdall 1y ago
I don't think the question would be whether the technology literally disappears entirely, only how important it is going forward. The metaverse is still technically here, but that doesn't mean it is impactful or worth near the investment. For LLMs, the architecture will be here and we know how to run them. If the tech hits a wall, though, and the usefulness doesn't balance well with the true cost of development and operation when VC money dries up, how many companies will still be building and running massive server farms for LLMs?
____mr____ 1y ago
If the tech is here to stay, my question is: how and why? The how: The projects for the new data centers and servers housing this tech are incredibly expensive to build and maintain. These also jack up the price of electricity in the neighborhoods and afaik the US electrical grid is extremely fragile and is already being pushed to its limit with the existing compute being used on AI. All of this for AI companies to not make a profit. The only case you could make would be to nationalize the companies and have them subsidized by taxes. But why?: This would require you to make a case that AI tools are useful enough to be sustained despite their massive costs and hard to quantify contribution to productivity. Is this really the case? I haven't really seen a productivity increase worth justifying the cost, and as soon as Anthropic tried to even remotely make a profit (or break even) power users instantly realized that the productivity is not really worth paying the actual compute required to do their tasks
- ttoinou 1y ago
  Do you need a measure or a quantification to do anything in life ? I don't wait for others benchmarks or others computing ROI factors to actually start using a technology and see it improves my workflow how and why? How : we'll always be able to run smaller models on consumer grade computers Why : most of the tasks humans need to do that computers couldn't do before, now can be improved with new AI. I fail to see how you can not see applications of this
Gazoche 1y ago
And, conversely, some don't care about the technology but want to ride the bubble and exit right before it pops.

rwmj 1y ago

The technology is there til the GPUs become obsolete, so about 3 years.

ttoinou 1y ago
Our current GPUs can last a decade or more, no ?
- rwmj 1y ago
  They (probably) won't physically fail, but they'll be obsolete compared to newer GPUs which will have more raw compute and lower power requirements.

_heimdall 1y ago

I don't think the question would be whether the technology literally disappears entirely, only how important it is going forward. The metaverse is still technically here, but that doesn't mean it is impactful or worth near the investment. For LLMs, the architecture will be here and we know how to run them. If the tech hits a wall, though, and the usefulness doesn't balance well with the true cost of development and operation when VC money dries up, how many companies will still be building and running massive server farms for LLMs?

____mr____ 1y ago

If the tech is here to stay, my question is: how and why? The how: The projects for the new data centers and servers housing this tech are incredibly expensive to build and maintain. These also jack up the price of electricity in the neighborhoods and afaik the US electrical grid is extremely fragile and is already being pushed to its limit with the existing compute being used on AI. All of this for AI companies to not make a profit. The only case you could make would be to nationalize the companies and have them subsidized by taxes. But why?: This would require you to make a case that AI tools are useful enough to be sustained despite their massive costs and hard to quantify contribution to productivity. Is this really the case? I haven't really seen a productivity increase worth justifying the cost, and as soon as Anthropic tried to even remotely make a profit (or break even) power users instantly realized that the productivity is not really worth paying the actual compute required to do their tasks

ttoinou 1y ago
Do you need a measure or a quantification to do anything in life ? I don't wait for others benchmarks or others computing ROI factors to actually start using a technology and see it improves my workflow how and why? How : we'll always be able to run smaller models on consumer grade computers Why : most of the tasks humans need to do that computers couldn't do before, now can be improved with new AI. I fail to see how you can not see applications of this

Gazoche 1y ago

And, conversely, some don't care about the technology but want to ride the bubble and exit right before it pops.