I don't think people really understand how powerful the next generation of AI models is going to be.Claude Mythos, when it comes out, will be the largest AI model ever made. After GPT3, AI companies stopped scaling up models because of the compute requirements. They just focused instead of improvements, like algorithmic improvements, building tooling around the models, improvements in techniques of training models, and reasoning improvements. Mythos will be the first generation of the big boys. I doubt it'll be cheap given all the compute it'll take to train and run it. There's no way these companies could charge the same rates they've been charging.They will come with all the improvements in technique learnt from the GPT3-GPT5 generation, as well as agentic improvements in tooling and be scaled up in size at a multiple from the generation we're at now.I think people are going to be really surprised, but they won't be able to afford it until compute costs come down drastically. Corporations will, though. I think when corporations start paying subscriptions to Myrthos there's going to be a scramble for hiring experts in prompting, because every hour their company is running the model, it'll cost. And they'll need people that can really efficiently use these big models.
>>108525647Don't care. I will still be waking up early to work a job I don't care about that I really should be getting paid more to do to afford living in spaces I don't like that cost 3x what they should while surrounded by people whose mind is perpetually absent from the real world.Until a human-like agi gets made, no one but sheckelberg and co and people whose jobs might be at risk will care about it.
>>108525647You're retarded for buying into hype before actual results are available.
Firstly, as anon above said, you're retarded for buying into hype before any results are known. Secondly, these models will be so much more costly that even Sam Altman, who has the most well-funded slop startup, decided it wasn't worth pursuing such a large model. These companies are all heading for the scrap heap, and none of the models so far have been shown to provide more benefits than their costs, aka they are failed technology.
>>108525647>expert in prompting>implement feature>don't make mistakes!
>>108525647>Powerful Who gives a fk.If it's not going to give us full luxury space communism then it's not worth increasing prices for energy, water, and muh PC parts.
>>108525647Can't wait to see how much more water and electricity they're gonna suck up so they can hallucinate even more advanced nonsense. Caaann't wait.
>>108525647Yeah, let's send those RAM prices even higher now.
Luddites will cope and seethe, as is tradition.
>>108525647how about you cut the marketing talk and specify actual benchmarks that you plan to run on the "big boy" models, which we could then replicate? quit yapping about how vaguely good they're gonna get, give us something to check if your words are worth anything substantial when the models are finally here. a track record of "the technology did get better when they threw billions at it, told you guys!!!" is worth nothing
>>108525647AGI in two weeks
>>108525958the specifics of these models is a secret. How can I tell you?Anthropic and OpenAI have disclosed that these models will be huge. They're not saying what they're doing exactly.The scaling law stands and hasn't shown been shown to break down. So when these scaled up models come out, based on the scaling law, and current capabilities, what do you think is going to happen? These models will be significantly more capable than Claude Opus 4.6https://arxiv.org/pdf/2001.08361
>>108526000Yeah, but knowing Anthropic it's gonna be $100/Mtok with a $1000 monthly subscription
>>108525647n4k4d4sh1
>>108525647>Trust me, this time it's real AGI, just two more weeks, just one more investment round of 100 billion dollars
Being larger and better doesn't impress me when we've reached peaks of diminishing returns.>it's 100x bigger and even less practical to run yourself!!This just doesn't excite me.
>>108526000The Chinese will reverse engineer it and then make it 32 times more efficient
>>108526000>the specifics of these models is a secret. How can I tell you?>They're not saying what they're doing exactly.I'm asking you what you plan to run on them to prove they're as capable as you think. Personally I'd hand it a formal specification and ask it to implement it but you do you
>>108525647>Claude Mythos, when it comes out, will be the largest AI model ever made*blocks your path*
>>108525647You guys are NEVER going to get to use the big boy models ever again, this is the end, after these they'll create even more expensive models, year after year, until it costs $10,000 to run it for a single minute.You will get half-assed small/medium distillations that are 1/100th as powerful. Compute will NEVER come down.You will never experience what the most powerful models are capable of, unless you get a job at a big AI company.
>>108525647i dont think next gen AIs will be any more powerful, we're probably already at beginning of diminishing returns now. I mean how much further can we push the transformer architecture?
>>108525647Why would you assume out of the blue that they have not yet hit the diminishing returns point in terms of how much compute they throw at it?Last time I checked, they run out of data to hoard. Not out of compute.Also there was no significant improvement in architecture. They were doing longer training.These are the 3 things that you make bigger -> get better inference results: compute, data, training time.Everything is maxed out already, data is the bottleneck and there's nowhere to get data from.They are also actively poisoning public datasets to throw competition behind. Each player assumes only he knows where the poison is, since he planted it. But there are many players. They will destroy the Internet with slop unless Age Verification laws would be implemented properly. Which is basically IDs for humans to tell them apart from bots.
>>108525647please stop telling me how i'm not prepared for something or how powerful something is going to be. you fucking chicken little faggots
AGI in two weeks sister
>>108525647>I don't think people really understand how powerful the next generation of AI models is going to be.So they're going to have much larger contexts? (That makes them a lot more expensive with standard AI architectures.)So they're going to avoid hallucinating lots of things that don't exist? (Some people distinguish between hallucinations and mirages; I don't give a shit. It's made shit up and lied about it.)So they're going to be possible to implement using a power budget of a few watts? (There's loads of applications that would be enabled by that!)I believe these are solvable problems. I also believe that Big AI can't solve them because they're fundamental architecture problems, and the cost of solving them is to obliterate their entire investment in hardware and datacenters, and drive them straight into bankruptcy.
>>108527914>AGI in two weeksCan those two weeks please be brought forward so they happen this century?
>>108526230Huh?
>>108527798Data isn't a bottleneck. Synthetic data can be generated.
Yeah dude, totally. You see, current AI is like a dot, but the next version is like a circle. mind=blown
>>108525790sounds like a personal problem, you should have voted in representatives that would protect your interests, my local politicians made it illegal for power companies to raise rates to subsidize data centers and the same for water utilities. also the pc part increase is artificial and greed driven by the manufacturers.
>>108525647TRUKEAIGODS WON, LUDDITES LOST
Remember that GPT3 was so intelligent that it had to go private to make mone- I mean preserve the human race.
>>108525647lewd that nipper
>>108529487Self recursion is poison dumbass
>>108529786b... bu... but only with AI for some reason! it's ok when humans do it.
>>108529802Yes, it matters when you're burning trillions trying to create the machine god
