3. Ai power is hungry and Hungary is getting.
You may have heard that Ai is hungry. But this reputation is due to a lot of electricity quantity in which the taxes of training these giant models are the amount of electricity, though the giant model is trained only every time.
What has changed is that these models are now using hundreds of millions of people every day. And when the use of a model takes less energy than a training, energy costs with this type of consumer numbers increase extensively.
For example, Chat GPT has 400 million weekly users. This becomes the fifthest website in the world after Instagram and beyond X.
So it is no surprise that tech companies are running to create new data centers in the desert and improve the power grid.
In fact, we have been in the dark of how much energy we take to accelerate because no major companies that build this technology have shared much information about it.
However, this is starting to change. Many of my colleagues spent working with researchers just to crush numbers for some open source version of this tech. (Check what they got.)
4. No one knows exactly how large language models work.
Certainly, we know how to build them. We know that they really do the way to work better. 1 on this list.
But how they do it is still a solution. It is as if these things have come from the outer space and scientists are standing outside and standing up to find out what they are really.
It is incredible to think that there is never a widespread market technology that is used by billions of people.
Why does it matter? Okay, unless we understand them better unless we know exactly what they can do and what they cannot do. We do not know how to control their behavior. We will not fully understand the deception.
5. AGI has no meaning.
Recently, it was a matter of Aigi, and the mainstream researchers were ashamed to bring it out. But since the AI has become better and much more profitable, serious people are happy to insist that they are about to create it. Whatever
AGI – or artificial general intelligence – means something like: AI which can accomplish human performance on extensive academic tasks.
But what does that mean? How do we measure performance? Which human? How many tasks limit? And performance on academic tasks is another way to say intelligence – so praise is circular anyway.
Basically, when people refer to Agi, they now only mean AI, but today we have better than what we have.
This is absolute faith in Ai’s progress. It has improved in the past, so it will continue to improve. But there is evidence of zero that it will actually end.
So where does it leave us? We are creating machines that are getting great in imitating some of the tasks of people, but this technology still has serious flaws. And we’re just knowing how it actually works.
How do I think of AI here: We have made machines like human behavior, but we have not removed the habit of imagining a human being like a human being. As a result, what AI can do and play in the wider culture wars between techno -optimists and techno -shakes.
It’s okay to be surprised by this technology. There are also doubts about many things that are said about it. There are many early days now, and it’s all in grip.
This story was actually published on AI in our weekly newsletter, algorithm. Sign up first to get such stories in your inbox Here.