- A human uses between 100W (naked human eating 2000kcal/day) to 10kW (first-world per capita energy consumption).
- Frontier models need something like 1-10 MW-years to train.
- Inference requires .1-1kW computers.
So it takes thousands of human-years to train a single model, but they run at around the same wall clock power consumption as a human. Depending on your personal opinion, they are also .1-1000x as a productive as the median human in how much useful work (or slop) they can produce per unit time.
What a depressing view of life. I don't expect him to take on some religious or philosophical view, but come on, how could you grow up somewhere wonderful, start a successful company with a lot of people you probably like and enjoy working with, have enough money to buy an island and still summarize life like that.
I prefer Richard Brandson's worldview. He's rich, but seeing the way he talks about his late wife and her memory warms my heart. I envy him for the human parts of his life, not just the success.
I see some folks here defending Altman because it was an off-the-cuff remark in front of a receptive audience. But why does this make the comment acceptable? Would you give me if I talked about eating babies, but defended myself by saying that I was speaking to a receptive audience?
Most charitably, it's a dumb thing to say. It compares two unrelated things if you see the value of human life to be more than just answering prompts. Less charitably, the argument is evil: if he was trying to make a sincere apples-to-apples comparison, it implies that he doesn't value human life beyond the labor his company can automate.
I can understand edgy teenagers making arguments like that on LessWrong forums, but Altman ought to know better. He either doesn't, or he sincerely believes what the comment implies.
This is a profound category error. What Altman reduces to a 20-year 'training' cycle fueled by 'energy' is what we, in the actual world, call life. It is a stunningly hollow perspective that uses the language of industrial output to describe the human experience. While he is likely being provocative to keep his product at the center of the cultural conversation, it probably exposes something about him.
I didn't read/hear it as reducing human life to 'training energy', but I don't like the comparison at the technical level.
Firstly, the math isn't even close. A human being consumes maybe 15 MWh of food energy from years 0 to 20. Modern frontier models take on the order of 100,000 MWh to train. It's a 10,000x difference. Furthermore, the human is actively doing 'inference' (living, acting, producing) during those 20 years of training and is also doings lots of non-brain stuff.
Besides the energy math, it's comparing apples-to-oranges. A human brain doesn't start out as a blank slate; it has billions of years of evolutionary priors for language and spatial reasoning that LLMs have to teach themselves from scratch, so this could explain why a human can do some things cheaper. Also, the learning material available to a human is inherently created to be easily ingested by a human brain, whereas a blank LLM needs to build the capacity to process that data.
Altman seems to hint at a comparison to the whole human evolution, but that seems unfair in the other direction, because humans and human evolution had to make discoveries from scratch and trial and error whereas LLMs get to ingest the final "good stuff". But either way you slice it, it's just not a good comparison, though not an 'inhuman' or immoral one.
I think this reveals a great deal about the thinking of the ruling elites.
The K shaped recovery phenomenon demonstrated that the economy can continue to thrive, when consumption by the lowest earners is replaced and concentrated by earners at the top. This demonstrated to the elites that actually, we don't need as many consumers to grow the economy, and that it's possible to redistribute wealth upward without losing growth.
These public comments just show that the elites are more and more comfortable making it explicit that there are too many "useless eaters" in their opinion, and that the change has been from considering just the Third World to be where these "useless eaters" are while still preserving an imperial core, to now considering everyone that isn't them, regardless of First or Third world, to be a useless eater.
Very dangerous thinking, but at least it's out in the open now.
They want to capture the entire value of everyone's labor and hoard it for themselves, and discard the people that produced it.
So he's comparing a human being to AI, finally showing what our AI overlords think of humanity: we're just wasteful resources to be replaced by more efficiency tools.
Elon Musk is perhaps the world’s most famous doom-monger and has repeatedly sounded the alarm about the possibility of super-smart machines wiping out humanity.
But Google founder Larry Page allegedly dismissed these fears as ‘speciesist’ during an argument at a Napa Valley party in 2015.
A top professor at the Massachusetts Institute of Technology (MIT) has claimed the two tech moguls clashed in a ‘long and spirited debate’ in the early hours of the morning.
In his book Life 3.0: Being Human In The Age of Artificial Intelligence, Max Tegmark wrote: ‘[Page’s] main concerns were that AI paranoia would delay the digital utopia and/or cause a military takeover of AI that would fall foul of Google’s “don’t be evil” slogan.
‘Elon kept pushing back and asked Larry to clarify details of his arguments, such as why he was so confident that digital life wouldn’t destroy everything we care about.
‘At times, Larry accused Elon of being “speciesist”: treating certain life forms as inferior just because they were silicon-based rather than carbon-based.’
The problem I see is that in our society, CEOs are chosen for their ability to convince that they can increase productivity. Not for their ability to improve the life of people.
Just like the paperclip AI issue, CEOs are optimising for arbitrary metrics, and they are really good at that (because we select them precisely for that).
So obviously, as soon as you start wondering about how competent a CEO is at talking about life, you're in for a treat. He obviously has no idea about life. He is just a successful paperclip production machine.
What scares me is that we select those people for their ability to convince that they will generate money, in the hope that they will actually do that, and then we value their opinion about completely unrelated topics.
You may as well ask a curling professional athlete what they think about the problem of AI and energy. Not that they necessarily will say something as dumb as Altman of course, but you wouldn't behave as if they were experts in the field of... you know... the impact of energy on humanity and life in general.
CEOs are a mix of scary, funny, innovative and naive persons. It's the first time an LLM is compared with a human in terms of energy. I will not comment the foolishness and superficiality of the quote. I would add that a human can meet another human and they make another human.
27 comments
[ 2.9 ms ] story [ 37.1 ms ] threadHe’s clearly saying “lots of important things consume energy” not “let’s replace humans with GPUs” or “humans are wasteful too”.
If Altman is to blame for anything, it’s that AI is a scissor-generator extraordinaire.
Unfortunately these companies are working to eliminate jobs, but not in any way making a path for a transition to a post-work society.
- A human uses between 100W (naked human eating 2000kcal/day) to 10kW (first-world per capita energy consumption).
- Frontier models need something like 1-10 MW-years to train.
- Inference requires .1-1kW computers.
So it takes thousands of human-years to train a single model, but they run at around the same wall clock power consumption as a human. Depending on your personal opinion, they are also .1-1000x as a productive as the median human in how much useful work (or slop) they can produce per unit time.
I prefer Richard Brandson's worldview. He's rich, but seeing the way he talks about his late wife and her memory warms my heart. I envy him for the human parts of his life, not just the success.
We only care about pelicans riding bicycles
Edit: Or perhaps more correctly, "less valuable human". Which is more appropriate?
Most charitably, it's a dumb thing to say. It compares two unrelated things if you see the value of human life to be more than just answering prompts. Less charitably, the argument is evil: if he was trying to make a sincere apples-to-apples comparison, it implies that he doesn't value human life beyond the labor his company can automate.
I can understand edgy teenagers making arguments like that on LessWrong forums, but Altman ought to know better. He either doesn't, or he sincerely believes what the comment implies.
Firstly, the math isn't even close. A human being consumes maybe 15 MWh of food energy from years 0 to 20. Modern frontier models take on the order of 100,000 MWh to train. It's a 10,000x difference. Furthermore, the human is actively doing 'inference' (living, acting, producing) during those 20 years of training and is also doings lots of non-brain stuff. Besides the energy math, it's comparing apples-to-oranges. A human brain doesn't start out as a blank slate; it has billions of years of evolutionary priors for language and spatial reasoning that LLMs have to teach themselves from scratch, so this could explain why a human can do some things cheaper. Also, the learning material available to a human is inherently created to be easily ingested by a human brain, whereas a blank LLM needs to build the capacity to process that data. Altman seems to hint at a comparison to the whole human evolution, but that seems unfair in the other direction, because humans and human evolution had to make discoveries from scratch and trial and error whereas LLMs get to ingest the final "good stuff". But either way you slice it, it's just not a good comparison, though not an 'inhuman' or immoral one.
The K shaped recovery phenomenon demonstrated that the economy can continue to thrive, when consumption by the lowest earners is replaced and concentrated by earners at the top. This demonstrated to the elites that actually, we don't need as many consumers to grow the economy, and that it's possible to redistribute wealth upward without losing growth.
These public comments just show that the elites are more and more comfortable making it explicit that there are too many "useless eaters" in their opinion, and that the change has been from considering just the Third World to be where these "useless eaters" are while still preserving an imperial core, to now considering everyone that isn't them, regardless of First or Third world, to be a useless eater.
Very dangerous thinking, but at least it's out in the open now.
They want to capture the entire value of everyone's labor and hoard it for themselves, and discard the people that produced it.
A human at rest used ~100Wh, up to 400Wh for an elite athlete under effort.
So 20 years at 200Wh (I'm being generous here) ends up being 35MW, still cheaper, and inference is still at under 200Wh!
In that light Altman saying things things like that is not really surprising. Contrary it only reinforces their desperation to me.
Context:
Elon Musk is perhaps the world’s most famous doom-monger and has repeatedly sounded the alarm about the possibility of super-smart machines wiping out humanity.
But Google founder Larry Page allegedly dismissed these fears as ‘speciesist’ during an argument at a Napa Valley party in 2015.
A top professor at the Massachusetts Institute of Technology (MIT) has claimed the two tech moguls clashed in a ‘long and spirited debate’ in the early hours of the morning.
In his book Life 3.0: Being Human In The Age of Artificial Intelligence, Max Tegmark wrote: ‘[Page’s] main concerns were that AI paranoia would delay the digital utopia and/or cause a military takeover of AI that would fall foul of Google’s “don’t be evil” slogan.
‘Elon kept pushing back and asked Larry to clarify details of his arguments, such as why he was so confident that digital life wouldn’t destroy everything we care about.
‘At times, Larry accused Elon of being “speciesist”: treating certain life forms as inferior just because they were silicon-based rather than carbon-based.’
(https://metro.co.uk/2018/05/02/elon-musks-fears-artificial-i...)
Just like the paperclip AI issue, CEOs are optimising for arbitrary metrics, and they are really good at that (because we select them precisely for that).
So obviously, as soon as you start wondering about how competent a CEO is at talking about life, you're in for a treat. He obviously has no idea about life. He is just a successful paperclip production machine.
What scares me is that we select those people for their ability to convince that they will generate money, in the hope that they will actually do that, and then we value their opinion about completely unrelated topics.
You may as well ask a curling professional athlete what they think about the problem of AI and energy. Not that they necessarily will say something as dumb as Altman of course, but you wouldn't behave as if they were experts in the field of... you know... the impact of energy on humanity and life in general.