Kurt Vonnegut wrote "Player Piano" in 1952. When I went to earn my engineering degree in the late 1980s we read and discussed it in a course about technology and its effect on society. I encourage people to read it. 1952!
What people do and what happens when they do it often does not work out as plan or intended. The reality we inhabit with the successes and failures is mostly a matter of luck. Most people who have been lucky believe they owe their station in life to hard work and insight. Mostly they deceive themselves and since they are deceived, the stories they tell and the beliefs they hold perpetuate this deception. I am guessing that what AI has done and will do is participate in this deception. If and when this becomes commonly believed to be true, it will be really painful for a lot of people and not the people who have become aware of the deception earlier than most.
This poisoning of AI reminds me of the scene in The matrix where they talk about why everything tastes like chicken because the AI in the story couldn't discern flavor. I can imagine a future perhaps a thousand years from now without cows and people studying history on the internet thinking they had four wheels and a bumper.
It's gotten to the point where people are using browser extensions to go back in time to before AI took over. The lack of trust in the internet feels far more permanent.
It's going to be a mixed bag. The cats have escaped the container and they will go feral and they will warm our hearts just like another human. bravo and well done. thank you for your time. form Oklahoma USA
Where is all the electricity supposed to come from for all this AI ? I get the feeling that the current large model AI is a sloppy data structure that uses too much hardware and electricity.
What AI will do is figure out a way to do things that used to be done by people. Instead of employing tens or hundreds or thousands of people, they will try to employ one computer. All the earnings of thousands of people will now go to a few dozen. That is only a good outcome for the few dozen. The thousands will correctly work to prevent that, and I will side with the thousands every time.
I think you're perpetuating a mistake that is made a lot about AI. AI isn't sentient and it doesn't have goals and it doesn't do things. It's a tool. So the problem isn't keeping AI's working for humans. It's keeping AI from being used as a tool by unscrupulous humans. It's a programming technique. So saying that you're trying to keep AI working for the benefit of humans is like saying you're trying to keep object-oriented programming working for the benefit of humans. That's not the issue. The issue is, who is using these programming techniques, what kinds of programs are they writing with them, what do those programs do, and are they permitting people to do things which really ought to be crimes. I know, science fiction is more fun. But in the real world, the problem with AI is that people use it as a mask and generate ads that are deceitful in a new and complex way. That should be where we're concerned.
I would like to share your optimism. But I can't guarantee that the AI's the public knows about are the extent of the technology. AI's (of which LLMs are a subset), may be tools, but in the non-linear non-predictable way in which nervous systems are "tools" of DNA. And expert systems most definitely do have "goals". They are baked into the reward functions used to train the "neural net" matrices.
32:40 "The hype bubble is bursting..." No, I think this is ill informed. He talks about the problem being a shortage of data, but if we take a step back, perhaps it is sufficient to build a model that is better than any human ever, or in fact the combined effects of all humans in sorting things out / being intelligent. The thing is, there is no single human nor the entire collection of all humans that will make real-time well-digested use of all data ever... So actually, training efficiency is what's important. So then consider the deepseek r1 model, which is competitive on many metrics with chatGPT 01 but trained for something like 1/10 the price. That cost may reflect a significant improvement in training efficiency, understood to mean the capacity to develop coherent and useful models per training set volume. The ability to develop coherent and useful models could be considered to approximately equate to intelligence, or at least a central aspect of it.
Deepseek thinks it is chatgpt. Most people think deepseek is just a focused layer on top of chatgpt. It would still take that much effort to make the model from scratch.
Kurt Vonnegut wrote "Player Piano" in 1952. When I went to earn my engineering degree in the late 1980s we read and discussed it in a course about technology and its effect on society. I encourage people to read it. 1952!
What people do and what happens when they do it often does not work out as plan or intended. The reality we inhabit with the successes and failures is mostly a matter of luck. Most people who have been lucky believe they owe their station in life to hard work and insight. Mostly they deceive themselves and since they are deceived, the stories they tell and the beliefs they hold perpetuate this deception. I am guessing that what AI has done and will do is participate in this deception. If and when this becomes commonly believed to be true, it will be really painful for a lot of people and not the people who have become aware of the deception earlier than most.
This poisoning of AI reminds me of the scene in The matrix where they talk about why everything tastes like chicken because the AI in the story couldn't discern flavor. I can imagine a future perhaps a thousand years from now without cows and people studying history on the internet thinking they had four wheels and a bumper.
It's gotten to the point where people are using browser extensions to go back in time to before AI took over. The lack of trust in the internet feels far more permanent.
It's going to be a mixed bag. The cats have escaped the container and they will go feral and they will warm our hearts just like another human. bravo and well done. thank you for your time. form Oklahoma USA
Imitation is the sincerest form of battery
Very clever! May I use this quote? If so, may I cite you as the author?
Where is all the electricity supposed to come from for all this AI ? I get the feeling that the current large model AI is a sloppy data structure that uses too much hardware and electricity.
What AI will do is figure out a way to do things that used to be done by people. Instead of employing tens or hundreds or thousands of people, they will try to employ one computer. All the earnings of thousands of people will now go to a few dozen. That is only a good outcome for the few dozen. The thousands will correctly work to prevent that, and I will side with the thousands every time.
I think you're perpetuating a mistake that is made a lot about AI. AI isn't sentient and it doesn't have goals and it doesn't do things. It's a tool. So the problem isn't keeping AI's working for humans. It's keeping AI from being used as a tool by unscrupulous humans.
It's a programming technique. So saying that you're trying to keep AI working for the benefit of humans is like saying you're trying to keep object-oriented programming working for the benefit of humans. That's not the issue. The issue is, who is using these programming techniques, what kinds of programs are they writing with them, what do those programs do, and are they permitting people to do things which really ought to be crimes.
I know, science fiction is more fun. But in the real world, the problem with AI is that people use it as a mask and generate ads that are deceitful in a new and complex way. That should be where we're concerned.
I would like to share your optimism.
But I can't guarantee that the AI's the public knows about are the extent of the technology.
AI's (of which LLMs are a subset), may be tools, but in the non-linear non-predictable way in which nervous systems are "tools" of DNA.
And expert systems most definitely do have "goals".
They are baked into the reward functions used to train the "neural net" matrices.
Should have called it Harlequin after the "hero" in Harlan Ellison's short story :-)
Interesting that this came out right before the half a trillion dollar drop in AI companies this week
32:40 "The hype bubble is bursting..."
No, I think this is ill informed. He talks about the problem being a shortage of data, but if we take a step back, perhaps it is sufficient to build a model that is better than any human ever, or in fact the combined effects of all humans in sorting things out / being intelligent. The thing is, there is no single human nor the entire collection of all humans that will make real-time well-digested use of all data ever... So actually, training efficiency is what's important. So then consider the deepseek r1 model, which is competitive on many metrics with chatGPT 01 but trained for something like 1/10 the price. That cost may reflect a significant improvement in training efficiency, understood to mean the capacity to develop coherent and useful models per training set volume. The ability to develop coherent and useful models could be considered to approximately equate to intelligence, or at least a central aspect of it.
Deepseek thinks it is chatgpt. Most people think deepseek is just a focused layer on top of chatgpt. It would still take that much effort to make the model from scratch.
That literally makes no sense.
This dealing with a very narrow use of AI, creative art reproduction. Utilitarian uses are far more common.
Apparent suicide? That is doubtful.
There is no "AI bubble". Don't be daft.
Wow, what a hostile interviewer.
Now you know what it feels like to step out of an echo chamber