>>105812401 (OP)

>>105812401 (OP)
>>105812406
Meme metrics.

>>105812406
>le extrapolashun
kys w/a crowbar and take your whole family with you

>>105812401 (OP)
>"Hey Alexa! How many R's are in the word strawberry?"
>"I'm sorry, I didn't understand, here's an online search for How many R's are in the world's straw bury."

We'll be ok for a while, anon.

>>105812406
Not a great record when 50% of the time it's not just wrong, it's confidently wrong. At least someone someone makes up an answer on a test they can tell you they made it up, the AI believes 100% that it did it correctly. Oh and the other 50% of the time is more than likely to contain at least one or two false gems.

>>105812401 (OP)
>just 2 more years

>ai will handle complex tasks in a few years
>tfw the said ai can't handle simple tasks today

>>105812401 (OP)
Aren't all actual measures showing that the improvements are slowing down drastically and that the only way to have them continue growing is by throwing endless resources on training them?

>>105812475
>Aren't all actual measures showing that the improvements are slowing down drastically and that the only way to have them continue growing is by throwing endless resources on training them?
yes, but the grifting must continue

>>105812406
>50%
So every time the AI does something it's a coin flip and half the times it's all wrong?

>>105812406
It's funny when they measure progress by making the models perform simple tasks in controlled environments as if brute forcing results through repetition means the model has suddenly acquired adaptive thinking.

>>105812434
This. Worry when market open reversals stop being a thing.

>>105812475
yeah but we are equally making society stupider

>>105812401 (OP)
>decided to give gemini-cli a go
>feed it with all our docs
>give it a detailed explanation about a new simple feature
>40 minutes to get it done
>doesn't work
I think we're fine

>>105812401 (OP)
>improving exponentially
lmao compared to what, 1975?

AI will be regulated to hell and back if it ever gets too smart and rich people start being affected by it.
When they start replacing upper managers, alphabet soup positions, even politicians and government workers, they will quickly put a stop to it and take a drill to its head to lobotomize it.

>>105812443
Based extrapolationist

>>105812401 (OP)
Rolling for new AI winter.

>>105812401 (OP)
If it can handle fucking, then we're very fucked.
Not holding my breath, tho.

Can't wait for the bubble to burst and all the AI faggots to kill themselves

>>105814390
If it actually leads to mass unemployment like AItards predict, the economy will take a hit. Without burning billions of dollars every day, the companies won't be able to keep these things online and cheap for consumers, so the AI bubble will burst.

>>105812401 (OP)
>Large Language Models Are Improving Exponentially
my dick remains unconvinced

>>105812406
how does this compare to the speed of solving problems with a literal coinflip instead then?

i reckon a traditional computer could crunch those numbers faster

>>105817252
>she doesn't know they can just create money ex nihilo

>>105817459
As Ukraine shows us, that strategy doesn't work too good when confronted with hard resource demands.

>>105817479
>randomly brings out Ukraine
found the brown

>>105816991
There is no bubble.

Despite years of work 90% of ai shit is completely useless

>>105812401 (OP)
but llms scale quadratically in size with their context length. and you need to fit it in vram.
they already started dividing AI into various "experts" so they could scale across many gpus. And the result is what it is. You could say that a company is already just a large group of moron "experts" so it's emulating real life, but also in real life there are underperforming companies that you can't invest in no matter how much you scale them up and replace humans.

>>105817683
>You could say that a company is already just a large group of moron "experts" so it's emulating real life
hearty kek

Just one more doubling of parameters and AGI is here. Just ignore that fact that the training data approaches all released information available and then some and that still isn't enough to teach it robust reasoning.

>>105817514
>found the brown

>>105812406
I don't doubt that the models will get better, even just by brute forcing countless answers, all financed by trillions from the taxpayer's money. But these extrapolations will never not be funny to me.

Also, I can't be the only one who feels like the best uses are in image and video generation, maybe coding as the models keep improving, rather than asking an LLM to tell you the meaning of life and getting some boring public service announcement type answer back. Yet people keep focusing on the latter.

>>105817964
Oh and maybe medicine as well, although you could probably do that without AI just by creating a program that cross-references symptoms and test results. Which I'm pretty sure has been a thing long before the AI craze.

>>105817964
>ask assistance for writing my literotica masterpiece
>proceeds to lay out steps in doing so only to abruptly stop and change the texts into something totally different
>suggests I should talk about something else
Tuck them!

>2 more weeks for AGI