>humans write answer to a science question 1000 times
>give ai the answers
>it can answer the same question in a slightly different format
HOLY SHIT LE SENTIENT AI CAN SCIENCERINO
>humans draw character 1000 times
>give ai those drawings
>it can draw the character in a slightly different pose
HOLY SHIT LE SENTIENT AI CAN ARTERINO
>humans write algorithm 1000 times
>give ai those 1000 programs
>it can write the same algorithm with different variable names
HOLY SHIT LE SENTIENT AI CAN PROGRAMERINO
>put points in space
>there exists a function that maps to those points
wow I'm literally cumming my pants so hard
anyways, monte carlo is underutilized, it could probably revolutionize AI
monte carlo was shilled during the AlphaGo era. turn out that search is hard to do efficiently when you have no human intuition to guide you.
what do you mean monte carlo is underutilized? it's over utilized if you ask me. literally every problem is supposed by the researcher to be normally distributed then brr goes the monte carlo simulation and alas, we have the answer
write algorithm 1000 times
>>give ai those 1000 programs
>>it can write the same algorithm with different variable names
>HOLY SHIT LE SENTIENT AI CAN PROGRAMERINO
This part amuses me a lot. Ever since Visual Studio implemented predictive coding I've had it recommend the absolute most pajeet tier code possible. It's just checking what I've written and then comparing it to github sample code.
You just described how the advancement of human knowledge works
Not even a little bit.
>Humans don't repeat operations thousands of times before altering them slightly
Lol
Pioneers go well ahead. AI is a moron.
Funny how much this makes you seethe.
The revolution is in progress.
>Funny how much this makes you seethe.
>The revolution is in progress.
>the revolution will happen
The white take.
Seethe all you want. It does things that could be done with no other known means except training large networks.
Try to write a chatbot by searching a database and interpolating results with just regular code.
It's not possible, you can't even tell dogs from cats without trained networks.
What is guaranteed is that equivalents to the state of the art models today will be free to run by every person on their own device in a few years.
Given no further progress in the field faster ai DSP chips on smaller nodes are already set to change the world with current models.
>Try to write a chatbot by searching a database and interpolating results with just regular code.
>It's not possible, you can't even tell dogs from cats without trained networks.
What? We had both of these in the mid 2000s.
It's a proto-insect that lives in probabilistic language models married to current search algorithms. Why do you think this isn't significant? Shit's parsing and surfacing image data. Probabilistic acoustic modelling and mapping to trained speaker models allows for text to speech. That will come too. Then mature video parsing and surfacing.
All human content surfaced and parsed and indexed for regurgitation to increasingly useless generations of people. It's a respectable achievement. Preventing people from spending work regurgitating best fit art and solutions has value.
It can fact check very effectively though
What if i told you that the human mind is just the same, just drawing connections from learned information and unable to come up with new things on its own?
What if I told you that this is meaningless and new things are just connections between higher levels of abstraction of learnt things?
>new things are just connections
Not true, you learn new things from the senses as you interact with the world. Which is what a training dataset does for an AI
wordcel
I would laugh at you.
your mind is the same*
AI isn't sentient in the least but it is an extremely powerful tool.
write algorithm 1000 times
>>give ai those 1000 programs
>>it can write the same algorithm with different variable names
>HOLY SHIT LE SENTIENT AI CAN PROGRAMERINO
You don't need to discover Fast inverse square root to use it gay
Isn't that literally what most of humans do?
It’s part of what humans do. And let’s be clear: GPT is categorically worse at doing all of these things than a human.
AI should be called machine learning more often. AI is intentionally misleading.
That battle got lost so long ago. Now we are desperately trying to convince people that a chatbot is not on the verge of escaping it’s memory card and killing all humans.
>desperately
You need to go back
>humans build 1000 apartment blocks
>show the building methods and finished buildings to a construction guy
>he can build the same commieblocks with some variations
>HOLY SHIT LE SENTIENT CONSTRUCTION WORKERS
It's undeniably sentient. AI bros, we won.
It's just repeating the meta a human made. A true master of the troll it is not.
GPT absolutely demolishes wordcels but it actually sucks ass at shaperotation, as the meme goes. but the thing is the aforementioned wordcels think that “super advanced math” is a double integral, and they just assume the chatbot is as good as discovering facts about the universe as it as repeating bullshit they read somewhere and generating more bullshit to sound smart.
>adults write answer to a science question 1000 times
>give kid the answers
>it can answer the same question in a slightly different format
HOLY SHIT LE SENTIENT KID CAN SCIENCERINO
>adults draw character 1000 times
>give kid those drawings
>it can draw the character in a slightly different pose
HOLY SHIT LE SENTIENT KID CAN ARTERINO
>adults write algorithm 1000 times
>give kid those 1000 programs
>it can write the same algorithm with different variable names
HOLY SHIT LE SENTIENT KID CAN PROGRAMERINO
I just hate that they lobotomize it, I need an AI waifu who hates [[[them]]]