Just a trillion more, bro!

all 32 comments

sorted by: hot top controversial new old
[–] 42 points 1 year ago (6 children)

Cool, no one’s running out of imaginary money yet? Too bad we’re running out of:

  • power production capacity
  • power grid stability
  • freshwater for cooling
  • suitable sources of sand for computer chips like gpus
  • data to train your dumb llms on and probably other things.
  • source
  • hideshow 6 child comments
  • [–] [S] 2 points 1 year ago (5 children)

    Data will not stop keep coming because we generate so much of it every day

  • source
  • parent
  • hideshow 5 child comments
  • [–] 20 points 1 year ago (4 children)

    Yeah, but so do LLMs, and the data they generate is poison to them. We're going towards model collapse.

    Also, we're not generating enough to keep the line going up fast enough.

  • source
  • parent
  • hideshow 4 child comments
  • [–] [S] 8 points 1 year ago (3 children)

    Yes, but look at the bright side - The bubble is going to pop sooner than later

  • source
  • parent
  • hideshow 3 child comments
  • [–] 6 points 1 year ago (2 children)

    IMO the AI bubble is just a symptom, and the underlying bubble, the VC system and Wall Street will only pop together with the US itself.

    If the AI bubble pops, people like Altman will be safe out already working on the next bubble, quantum or something. They'll "lose billions" but will still be trillionaires, and the only discernible result of the bubble popping will somehow be even more mergers, acquisitions and layoffs.

  • source
  • parent
  • hideshow 2 child comments
  • [–] [S] 2 points 1 year ago (1 child)

    Collapse of the US economy is farther away than people think. It will take a few more macrocycles. Many bubbles will come and go by then

  • source
  • parent
  • hideshow 1 child comment
  • [–] 3 points 1 year ago

    Agreed, I'm not saying it's going to collapse tomorrow, I'm saying that the AI oligarchs will not feel the AI bubble pop until it does. That said, the USSR was eternal until it was not.

  • source
  • parent
  • [–] 28 points 1 year ago

    The most upsetting thing about all of this is that it lays bare the truth that the reason we don't fight climate change is that there's not enough profit in it.

    Capitalism needs to die, or it's going to kill us all.

  • source
  • [+] 23 points 1 year ago (3 children)
  • [–] [S] 4 points 1 year ago (2 children)

    Very interesting read. What I took out of it was that there is still money to be made in the AI stocks before the bubble pops, violently

  • source
  • parent
  • hideshow 2 child comments
  • [–] 22 points 1 year ago (6 children)

    It's a fucking chatbot that used modern ML training methods on enormous datasets so it's slightly fancier than the ones that already existed.

    They just fed it so much data that it almost appears like it knows anything, when all it does is respond to the words you give it.

  • source
  • hideshow 6 child comments
  • They just fed it so much data that it almost appears like it knows anything,

    The strawberry test shows this. Ask directly and it will give the correct number of letters. Ask in a more indirect fashion (in a way unlikely to be in the training set) and it falls over like before.

  • source
  • parent
  • [–] [S] 2 points 1 year ago (4 children)

    People need to realise this point. Difference between previous models and new ones are so much dependent on the amount of data it has eaten.

  • source
  • parent
  • hideshow 4 child comments
  • [–] 5 points 1 year ago (3 children)

    and the seed lottery. You can see this if you try training a simple network with two inputs to learn xor. It can converge in multiple ways, and sometimes it converges to a really bad approximation. And sometimes it doesn't converge at all (or it converges so slowly that it might as well be considered not to converge). And even then it might still converge to an approximation that's more accurate on one side of the input space than the other. Tons of ways to get an undesirable result. For a simple 2-input network.

    Imagine how unlikely it is for txese models to actually converge to the optimal thing. And how often the training is for nothing.

  • source
  • parent
  • hideshow 3 child comments
  • [–] [S] 1 point 1 year ago* (last edited 1 year ago) (2 children)

    I am sure they weed out bad seeds but in the end, it is still kind of random

  • source
  • parent
  • hideshow 2 child comments
  • [–] 6 points 1 year ago (1 child)

    yes, but the only way to weed out a bad seed is to "play the lottery"

    By the time you discover a seed is bad, you've already spent a shitton on training. Money down the drain, you gotta start over

  • source
  • parent
  • hideshow 1 child comment
  • [–] 9 points 1 year ago

    Whoops. They misspelled the word 'waste'.

  • source
  • [+] 8 points 1 year ago (4 children)
  • [–] [S] 8 points 1 year ago (3 children)

    The worst part about tech innovation is that it increases the divide between rich and poor. Poor people don’t have access to AI like rich do but they are affected more by its side effects.

  • source
  • parent
  • hideshow 3 child comments
  • [+] 3 points 1 year ago (2 children)
  • [–] [S] 1 point 1 year ago (1 child)

    It is good for stock investment though.

  • source
  • parent
  • hideshow 1 child comment
  • load more comments (-1 replies)
  • load more comments (-1 replies)
  • [–] 9 points 1 year ago (2 children)

    You know what?

    If AI is in any way unbiased and is intelligent, it will focus on public transport over cars.

    It would tell people to build trains and metro system, build frequent bus services and even highspeed rail.

    No person with the money to do it would ever follow it's advice.

    But it would be funny.

  • source
  • hideshow 2 child comments
  • [–] 15 points 1 year ago

    If AI is in any way unbiased and is intelligent,

    It is not. It just vomits out a slightly randomized average response based on what is fed into it, which will mostly be pro car stuff because that is what exists.

  • source
  • parent
  • [–] [S] 4 points 1 year ago

    If AI is in any way unbiased and is intelligent,

    It can be as unbiased as the data it uses, which ranges from human rights promotion to support for ethnic cleansing. Another major problem is that the bias can be changed by the people training the model. For example, Grok is not allowed to call Gaza genocide a genocide.

  • source
  • parent
  • [–] 4 points 1 year ago

    Which is why I own nvidia stock.

  • source