If revenue remains exponential, OpenAI will be the most profitable business on earth
Is this what's being spent on training or how much it would take to train an equivalent model over the years? If its the former that could also mean companies are just training less models.
>>108739089
>>108739089How terrible.
>if revenue remains exponentialThere's zero reason to expect this, it's absurd to think their growth rate will be maintained. OpenAI is not well positioned. Plenty of companies have demonstrated capabilities of generating cutting edge models and Chinese open source models aren't far behind. AI is going to be power and hardware constrained which openAI doesn't have any advantages for.I expect Anthropic to beat openAI
>>108739089>projections
>>108742015The US literally has 10x more compute than China and the gap is only getting wider. The reality is just that everyone wants to use American AI, not chinese, because the Chinese models are at least 3 generations behind when you actually use them. They're about comparable to Grok, but at least grok 4.3 actually has a working 1 million context (somewhat).
>>108742015The US and China will be siphoning money from the rest of the world with it, much more so than websites like Facebook, YouTube, or Google did. Though china will only get a much smaller pie.
revenue != profits
>>108739089>projectionsHAHAHAHAHAHAH
>>108739089>If revenue remains exponentialA thing that never happens.
at least it's sucking all the covid money out of the pockets of investors and into the economy, even if it's just data centers
>>108740017I like when the lines go up
>>108739089>projections>estimates>always in 2-3 years, never this yearjeetanon, I estimate I'll make one bazillion in 2027, if you give me a loan of 1 million today, I'll pay you back with 100% interest.proof? I have a bar graph!
>>108742109>The US literally has 10x more compute than China The non-existent stargate project doesn't count. >everyone wants to use American AIAs long as they're heavily subsidised by venture capitalists, NOT when they're at full token price.>at least 3 generations behindDoesn't mean much when model improvement is rapidly approaching diminishing returns.
>>108742015It's almost as if an ai companies projected profits are directly linked to the amount of copywrite violations they commit.
>>108739089My projections suggest OpenAI's training costs, and even operating costs, will be $0 by the 30s.
>>108739089>if revenue remains exponential Fool. The one certainty is that revenue will not remain exponential.
>>108739089>if INCOME LINE go UP!>and COSTS LINE go DOWN!!>we gonna LOSE LESS, by the DAY!!!
>>108744239Yes, it will be training itself by then.
>>108739089Projections thread? Projections thread! Yay! Dumping what I got.
>>108744342I heckin love projections!
>>108742109Meanwhile China is cranking out respectable models with good token efficiency
>>108744352
>>108744359I'm collecting projections, please find it in your heart to donate!
>>108744369We're actually a few projections short, ironically enough. I thought I had more projection related memes. Oh well... Maybe I saved them in the wrong folder. Will post if I find them.
>>108739089The one I sent you was a little bit of a day off.
so what'll be the magic bullet that suddenly stops the costs from going exponential lol
>>108739089Elon will his case though
the next break through will be organic AIThey'll merge lab grown brain cells with the AI
if revenue continues to grow exponentially and expenses stay flat, AI will become incredibly profitable in 5-10 years
>>108739089>If revenue remains exponentialAnon, there is no revenue , it's just the same cash moving between a select few parties
>>108744171You're joking, but investors are some of the dumbest motherfuckers alive
>>108744709Musk is just throwing shit at the wall in hope someone mistakes him as being still relevant. Even he knew Google was always gonna be the Final Victor. Which makes his now pissing cash up his own xAiHole, to be at best a bad third in a 1-horse race, so much the dumber.
>>108744369>>108744359>>108744352at birth my son was 51cm tall and weighed 3.3kg (20"/7.3lbs for the yanks)at 1 month, he is now 55cm tall and weighs 4.5kg (21.7"/10lbs)if these trends continue, at age 18 he will be 7.7m tall and weigh 263kg (25'3"/580lbs)
>>108744936The height might be a bit exaggerated, but the weight is spot on.
>>108742109(You) might think about Chinese models that way, but US AI companies don't seem to agree.
>>108745098Well isn't that just ironic
>>108739089>worried about training costs>take the attitude that you're supposed to know that already>don't spend money on training>???>profit
>>108740017fpbp /threadIt's also hilarious how even though everyone knew how fucked Itanium was as soon as it came out even in 2005 there were still retards getting their projections wrong by X00% kek
>>108745098This doesn't prove anything other than them hoovering up tons of data for training. Any text related to asking what model you are in chinese is going to be polluted.If this was specific proof of them distilling DeepSeek they would have to ask it in Chinese what model it is and then train against that answer. What exactly is the point of that and why would they do that?
>>108739089it's ok when corporations pirate shit
>>108744936>my daughter 99th percentile height whole first 2 years of life>everyone: oh wow shes gonna be a 6' model>nearly the shortest girl in class nowkek
>>108742109This is cope. I've used the Chinese models and the Mistral models and they are as goodd as the American ones.
