Enterprises in for a shock when they realize power and cooling demands of AI
theregister.com
theregister.com
All the clouds offer sovereignty controls now IIRC. If your data is so sensitive that it can't even leave the building, then you're probably going to have a bad time using it with your own home-grown AI solutions due to hallucinations.
Why are homegrown hallucinations worse than cloud hallucinations?
I don't think anyone apart from a FAANG-like corp really stands a chance in keeping up with SOTA.
An intern AI doesn’t need to be restricted in the same way.
Heck, that MacBook probably consumes less power with all 24 cores blazing than the 1-core computer did just to check email. Modern notebook CPUs do a really good job of leaving cores that aren't currently needed in power save mode. That's what makes it possible for the battery to last for a full workday. Despite possibly also having less capacity than typical laptop batteries from 25 years ago.
What people mean when they say "AGI" is something different.
what do they mean?
In humans ? /s
Fun read!
It is because AI will help us reduce CO2 output. /s
And before anyone says it can't grow forever, it's unsustainable, etc... you're not thinking big enough. Just google kardashev scales.
If your point is that "they used to say x, now they say y", my rebuttal is that 1) the exact same person/people? 2) it's common for people to be mistaken and then learn something new, and change their opinion. It might not happen in the middle of an internet argument, but it does happen, a lot.
log((mass universe * c^2)/(20 TW * 1 year)) / log(1.025) = 4600 years for 2.5% power consumption growth to exceed the rest mass of the universe per year: http://www.wolframalpha.com/input/?i=log%28%28mass%20univers...
Even if we had a warp drive (which may be impossible), we'd have to invent a free energy machine to sustain 2.5% growth for as long into the future as the earliest Egyptian pyramid construction was into the past, and also an infinite dark energy source to keep the universe from then collapsing into a black hole.
So it seems to me, given that much of life outside that distracting cellphone in your pocket is the same in terms of amenities affordances and available goods as it was in the 1950s, that there ought to be some efficiency gains made along the 75 years since. Then again maybe those efficiency gains are swallowed by more people globally being elevated from that great depression sort of desperate life we saw in this country in the 1930s and elsewhere today globally, into the prototypical 1950s lifestyle where a job begets housing, transport in some form, and the modern experience of stocked grocery stores of all goods instead of having to homestead or contribute to a village system for limited goods.
I'm sure I could go on and on with this for much longer. It's easy to not see our progress over time even though it's tremendous.
Exactly. Any society that seriously adopted it and deliberately deindustralised would eventually just get taken over by a more industrialised neighbour.
Waste? Cloud providers are super expensive and can't generally justify this expense with the benefits of outsourcing.
Yes though the cost breakdown has traditionally been large upfront development costs and low to moderate running costs. This time around the running costs are astronomical and Moore's law ain't what it used to be.
This is most R&D. You research, build a prototype, bring it to market, and it only hits profitability at volume.
Point is, we can look at the situation, and recognize "that can't go on forever". And then we can say so, sometimes in public places. And there's nothing wrong with that.
*Localizing Power Usage*: By moving some computational tasks to local systems (on-premises), you could reduce the reliance on large, centralized data centers, which are often constrained by power and cooling limitations. This would also reduce the need for extensive retrofitting of data centers to support higher power densities.
*Efficiency Gains*: Decentralized computation, especially for non-essential services, could be more energy-efficient on a per-unit basis. Local infrastructure might allow for better optimization of energy use. Smaller, distributed systems could have specialized power and cooling needs that are more manageable and tailored to the task at hand.
*Reduced Data Transmission and Latency*: With computing spread out across local or edge-based facilities, energy costs could be further reduced by minimizing the need for long-distance data transmission, which itself requires significant energy, especially with AI workloads.