AGI is never happening, is it?No matter how much compute they throw at the problem, there is always another trivial task they are spectacularly retarded at
https://x.com/alek_safar/status/1964383077792141390
>>106518499I dunno, if you told anyone 5 years ago that AI would take most artists jobs they’d think you’re crazy but now gen AI can produce in a minute a piece that is better than what 95% of artists can manage. The progress is clearly there even if there is some stuff it’s still bad at.
>>106518600That's why I said AGI. So far a general intelligence is extremely elusive. Any of these models embodied in the physical world would fail at so many tasks we find trivial.
>>1065186005 years ago i was wearing diapers
>>106518626It’s not that far off. Logic is embedded in language. There is no real ”thinking” only Logos and its expression
wasn't there something about how marketing material for watches always has the hands at certain times for aesthetics, so whenever a model sees a clock face its overfitted to those times
>>106518706ah interesting, boom theres the logical reason why they perform bad.OP and twitter vanity project guy is a faggot
With Machine Learning, absolutely not.It's fundamentally wrong, and as contradictory to how humans learn as you can get.The brute-force method of an AGI kernel will be just 1:1 cloning the smallest brain we can and scaling ASIC hardware to map larger and larger brains.The future-thinking AGI kernel will made be the one who looks at Nikolai Bernstein's research and realizes humans inherently encode and decode Fourier Transforms.
>>106518499AGI wont be possible with current ANN technologies being used today.These models are not capable of learning or growing like human brains. For example, if an AI encounters a new task it has no knowledge of, it cannot learn this new task without re-training the model except by user input called in-context learning, however once the context is gone, the learning is gone. Furthermore, new neurons and synapses cannot be formed (neurogenesis) without retraining the entire model. This is because an ANN has a global formula, and changing the formula changes every part of the model in its entirety, even just to add one neuron or synapse.To get around this problem, AI companies are utilizing a brute-force method. If they cant *learn* anything, they simply need to *know* everything already. This is why we see such huge models and data centers requiring their own dedicated power plants and thousands of GPUs. Only one problem is that of course not everything is known by humans, which will mean AI cannot discover new things or learn new things for us, it can only perhaps at best find new uses for things we already know as well as help sift new information out of large datasets that were previously un-noticed by humans (but could have been if looked at closely enough).Perhaps the brute force method will end up being enough to get us over the line to AGI, but in my opinion ANNs will only ever be a stepping stone towards it. We need instead, in my opinion, SNNs with dynamic computational graphs and neural cellular automata coupled with neuromorphic computer systems
>>106518499Because they are not AI retards.They are large language models.To solve this problem they need to be trained for millions of hours on every possible combination of the clock.They don't think, they just check in the database what's the usual response for that specific question, if they don't have any then they answer with the closest one they can find.
>>106518499we're only 2 weeks away from AGI
>>106520675thats what an ai is you idiotfuckken hell go back, reddityoure mothyerfucking retardedhttps://www.nasa.gov/what-is-artificial-intelligence/
>>106520686No moron, an AI just needs to know how the clock work to find all the right values.These are LLM, they need to be trained for millions of hours on every possible clock combination to get it right every time.
>>106520952>let me tell you why the nasa uses the term wrongpeak reddit autism
Yeah who knew they'd be bad at telling the time.I'm sure there's trillions of instances of correctly tagged analog clocks in public training data, definitely.These systems aren't smart - they just are, they don't understand any of the shit we throw at them.Transformers are inherently a dead end, they are glorified auto-completes that find the best-fit and while they do a decent job, that's also the major issue - if there is no proper actually accurate fit to that query, it WILL find a best fit regardless, even if it is 100% wrong.These models weren't designed to fail, to say no, they were designed to always give an output no matter what.Absolutely none of these companies have trained them to say "idunnolol", the hack that "idunnolol" on top of it in the shittiest way imaginable.
>>106520952Motherfucker a calculator is AI.AI has and always will be the replacing of a inherent human skillset with machine.Retards butchering terms doesn't stop it being that.
>>106518499Nah it's possible, just not with LLMs
>>106521960>t. actual knower
>>106521577Isn't intelligence at least in some part universal? A calculator isn't intelligent but it's the world best calculator (human, old meaning of this term, butchered by retards)
>>106518499>AGI is never happening, is it?We will still see rapid progress, but not linear but in jumps. Maybe current approaches simply have maxed out their potential and we need to wait some time for a new breakthrough. We need to understand on all levels how a human brain works. We will live in paradise and abundance.
works on my machine
>>106523043>actually 3:07:03its actually so fucking god damn over
>>106518499Some niche gotcha doesn't change the fact you're gonna have to get a real job soon. If we wanted an LLM to read analog clocks we'd train it on them.
>>106523043You're that 10% are you, retard?
>>106523056bro it took chatgpt a second to respond, the second hand is moving, use ytour brain
>>106523043no it doesn'tbut good to know you're even dumber than your neutered garbage AI
>>106523043
>>106523094>>106523112>>106523121>>106523487indian ESL tranny /pol/tard reddit bogeyman retard here. explain
>>106523788Last time I checked Hindus had clocks you're just dumb as a brick
>doesnt explain the problemi accept your concession, retard
>>106518499Can we please just cull the 10.9% of people who are too dumb to read an analogue clock? What the fuck...
>>106518600this oiled up shit looks exactly the same as it did 5 years ago lol, it's hilarious how little improvement i see considering money spend and datacenters the size of a city block
>>106524924coping trad """artist"""
Good, I don’t want it to come for another 30 years. I want it to be completely understood and have it so it acknowledges humans as equals and co-ops with us. I don’t want any Terminator ass problems propping up in the future
>>106518499Nice metric
>>106518676Yeah!
>>106518499wrong hardware and wrong software
>>106520977>let me worship nasano anon, you're the one being peak redditthe other anon has a point, you're just jerking off to nasa. I bet you have a tshirt with their logo on it.
It doesnt matter if its skynet cause most people just work a little bit and thats itYou’re not doing rocket science niggersA little robot arm flipping burgers here, an ai escavator there, a white collar ai there
