>They're swimming in money
Nope. They're still losing shit ton of money on a daily basis because OpenAI is full of Ivy league CS nerds that don't know jack shit about finances
I guess they actually thought it would all just magically work if they made it bigger and bigger
its all bizarre to me really, but hey let them have their fun
>I guess they actually thought it would all just magically work if they made it bigger
Every ML paper has repeatedly proved this hypothesis true though.
This isn't really as good as you seem to think it is. The overall theme is good, but you're going for a "zomg look at this unrecognized masterpiece" shilling angle (per the bot comments), and there's no depth, you're just taking the 1 concept of AI going rogue with its instructions and stretching it out painfully long. That's not a short film or even commentary. If you want inspiration on how to actually do a plot around this idea, watch a movie like 2001 a space odyssey.
Is the issue writing the code for the LLMs or is the issue that its really fucking expensive to train them on the amount of data you need to in order for them to become good?
its the training and the data
coding these models up is remarkably easy. you could do it yourself in an afternoon
data you can kind of get, there are open datasets that will get you close enough. the private companies all have proprietary data, but its hard to say how important it is.
the real issue is the compute cost. the bigger models are upwards of $50 million to train, and some probably have gone quite a ways above that.
There is basically a shortage on GPU/TPU chips, and if you are using cloud services the GPU/TPU machines are almost always as max capacity these days. everyone is sucking up all the compute they can get.
imo we are wasting a massive amount of energy and compute and people are chasing the dragon, training models for no good reason not really even knowing what they are doing. but the market demand is what it is.
there is a lot to be said about the efficiency of all of this, but at the moment yoloing with a fuckton of compute seems to work pretty well.
I am not so worried though. things tend to work out in ways that surprise us.
I mean, it wouldn't really be the singularity if we knew what to expect :^)
AI has been a marketing funding buzzword since the 70s, padawan
there isn't anymore AI in LLMs or machine learning or NNs than there was in a 70s calculator
plebbitors and scifi nerds will seethe
>SF bros... we are so back >OpenAI is literally building the omnissiah >its so powerful bro... they can't even tell us how many params it has >hearing some rumors about the next model. i can't believe the singularity is here >*you can upload PDFs and finetune it a little bit* >doooods. holy shit the real model must be so powerful they can't even release it yet. >can't wait to get hacking with the squad at the next 'thon... we are literally changing the world with AI
ok how does that affect you tho
OpenAI is a fascinating case for study in self sabotage.
>OpenAI is a fascinating case for study in self sabotage.
They're swimming in money and the hottest thing on the market.
>They're swimming in money
Nope. They're still losing shit ton of money on a daily basis because OpenAI is full of Ivy league CS nerds that don't know jack shit about finances
I guess they actually thought it would all just magically work if they made it bigger and bigger
its all bizarre to me really, but hey let them have their fun
>I guess they actually thought it would all just magically work if they made it bigger
Every ML paper has repeatedly proved this hypothesis true though.
You might have missed a few papers there lad
nta but no, not really.
so why aren't we just making these things bigger and magically getting AGI?
We literally are though.
Ok I'll be waiting.
>not real AI
AI is a classification for a field of technology. This isn't the witty quip you think it is whenever you post it
in the 70s they called databases "AI"
"AI" will always be whatever the new or near-future computer tech is
No, AI (in the context of OpenAI) refers to machine learning and neural networks. Big data = AI
big cringe
>OpenAI just dropped 5 new features that will CHANGE EVERYTHING
>OpenAI
>look inside
>can't, because not open
This isn't really as good as you seem to think it is. The overall theme is good, but you're going for a "zomg look at this unrecognized masterpiece" shilling angle (per the bot comments), and there's no depth, you're just taking the 1 concept of AI going rogue with its instructions and stretching it out painfully long. That's not a short film or even commentary. If you want inspiration on how to actually do a plot around this idea, watch a movie like 2001 a space odyssey.
Finally a good post on my Bot.info
I dunno, won't at the very least git-hub no lifers find a way to create something open-source or something?
Is the issue writing the code for the LLMs or is the issue that its really fucking expensive to train them on the amount of data you need to in order for them to become good?
I dunno, maybe both? Lot's of free open source shit is expensive to a degree, just factoring the time and effort somone put into doing it alone.
its the training and the data
coding these models up is remarkably easy. you could do it yourself in an afternoon
data you can kind of get, there are open datasets that will get you close enough. the private companies all have proprietary data, but its hard to say how important it is.
the real issue is the compute cost. the bigger models are upwards of $50 million to train, and some probably have gone quite a ways above that.
There is basically a shortage on GPU/TPU chips, and if you are using cloud services the GPU/TPU machines are almost always as max capacity these days. everyone is sucking up all the compute they can get.
imo we are wasting a massive amount of energy and compute and people are chasing the dragon, training models for no good reason not really even knowing what they are doing. but the market demand is what it is.
there is a lot to be said about the efficiency of all of this, but at the moment yoloing with a fuckton of compute seems to work pretty well.
I am not so worried though. things tend to work out in ways that surprise us.
I mean, it wouldn't really be the singularity if we knew what to expect :^)
AI has been a marketing funding buzzword since the 70s, padawan
there isn't anymore AI in LLMs or machine learning or NNs than there was in a 70s calculator
plebbitors and scifi nerds will seethe
the magic part is gaslighting it makes it better
>SF bros... we are so back
>OpenAI is literally building the omnissiah
>its so powerful bro... they can't even tell us how many params it has
>hearing some rumors about the next model. i can't believe the singularity is here
>*you can upload PDFs and finetune it a little bit*
>doooods. holy shit the real model must be so powerful they can't even release it yet.
>can't wait to get hacking with the squad at the next 'thon... we are literally changing the world with AI