The driving example wasn't meant to say you need a human level AI to drive a vehicle, just that I believe you do if you want it to drive just like a person. That's not a requirement, and as you point out, also not ideal. It was just an example to show the difference that exists.
IBM defines AGI: "Strong AI aims to create intelligent machines that are indistinguishable from the human mind. But just like a child, the AI machine would have to learn through input and experiences, constantly progressing and advancing its abilities over time."
But I do think there are many definitions and what may or may not be needed to get there. We can also accept that current AI doesn't need to be AGI to be very useful.
As for new stuff... does there exist an example of something novel or unique that came from nothing?
Everything we do at some point did not exist. Language, music, maths, science, philosophy, society, art, technology etc. At some point someone was the first to start down the path.
I get what you are saying, and again I'm inclined to somewhat agree.. how much of it is truly unique and how much of it is just remixing.. and since we live in a physical world we can't help be influenced by the birds and rocks and stars so maybe it is all just remixing?
But General Relativity for example isn't just a remixing of previous theories of gravity, it is a completely different way of understanding reality. Sure it requires knowledge of maths and physics and observable data, but a LLM could never come up with it.
__________________ Uncertainty is an uncomfortable position.
But certainty is an absurd one.
Everything we do at some point did not exist. Language, music, maths, science, philosophy, society, art, technology etc. At some point someone was the first to start down the path.
I get what you are saying, and again I'm inclined to somewhat agree.. how much of it is truly unique and how much of it is just remixing.. and since we live in a physical world we can't help be influenced by the birds and rocks and stars so maybe it is all just remixing?
But General Relativity for example isn't just a remixing of previous theories of gravity, it is a completely different way of understanding reality. Sure it requires knowledge of maths and physics and observable data, but a LLM could never come up with it.
I would agree... in their current state, an LLM could not come up with Relativity, which is basically the gold standard of human ingenuity and scientific creativity. I suppose my question would be: what is an LLM in 5 years?
The jump from GPT3 to what we'll have shortly where it can chat, "see," and "hear" all at the same time practically renders it a whole new thing. If GPT5 is the jump many feel it could be, it will open the floodgates further. I feel like it's hard to say what an LLM can and can't do because we don't know what an LLM in 5 years even is.
I think of the early machine learning videos where a computer was able to play the arcade game Defender indefinitely. It had an environment where it could play, lose, and learn rapidly, eventually figuring out a strategy that would allow it to survive. It is an environment of discovery, not just being fed/trained on known information.
Could there be an environment like that for AI to "discover" something like the theory of relativity? Possibly. It would require being able to run precise experiments and analyze results along with the application of already known data. It needs a way to rank theories , toss bad ones, and refine good ones. It's basically how genetic algorithms work now.
As Russic said, as AI begins to be able to interact more an more with the real world I think this becomes increasingly possible.
The thing I always think about when comparing humans with AI is every human starts at 0 at birth, and has to start the (relatively) slow process of gathering knowledge/experience to be able to discover new things. Then all of it is gone at death. Obviously not quite gone, but there is no other human with the same set of knowledge and experience, and the next human to be born has to start at 0 again.
With AI, it does not die. New models may have to be retrained, but they could theoretically live forever, continually learning. There may be a time when a new AI can simply ingest the models of previous AIs, and essentially pick up where the last AI left off. Regardless, it's a big advantage.
There's interdependence between the "hardware" and "software", babies are born with some instincts and hardware that's tuned to be receptive to the experiences the baby will experience, but yeah humans do start pretty close to zero even compared to some other species, because a lot of humans' advantage is in their social structure, which exists and evolves separately (though still influencing the biological evolution).
Yeah in theory an AI could just keep going.
__________________ Uncertainty is an uncomfortable position.
But certainty is an absurd one.
I think of the early machine learning videos where a computer was able to play the arcade game Defender indefinitely. It had an environment where it could play, lose, and learn rapidly, eventually figuring out a strategy that would allow it to survive. It is an environment of discovery, not just being fed/trained on known information.
Could there be an environment like that for AI to "discover" something like the theory of relativity? Possibly. It would require being able to run precise experiments and analyze results along with the application of already known data. It needs a way to rank theories , toss bad ones, and refine good ones. It's basically how genetic algorithms work now.
As Russic said, as AI begins to be able to interact more an more with the real world I think this becomes increasingly possible.
The thing I always think about when comparing humans with AI is every human starts at 0 at birth, and has to start the (relatively) slow process of gathering knowledge/experience to be able to discover new things. Then all of it is gone at death. Obviously not quite gone, but there is no other human with the same set of knowledge and experience, and the next human to be born has to start at 0 again.
With AI, it does not die. New models may have to be retrained, but they could theoretically live forever, continually learning. There may be a time when a new AI can simply ingest the models of previous AIs, and essentially pick up where the last AI left off. Regardless, it's a big advantage.
That sounds like the Q-Star training that OpenAI has been using (maybe? I can't recall where they are with it). It more or less explores an environment and tries a bunch of things, ranking what works best. Pretty sure we'll have to move onto that because I've heard we'll run out of human content to train it on by 2026.
This is where the "but it can't think" gets murkier for me. If an AI learns to play defender by trying 10,000 strategies and ranking them, is that not thinking? Perhaps it's not... does it matter though? Is the goal to think or rock ass at Defender?
If you have a manufacturing business and your AI assistant comes up with a market that may require your stuff that you'd never considered, that's a straight win. Does it matter if it came up with that idea in a flash of brilliance or it did some pattern matching on 5 million potential scenarios like Dr. Strange?
One nuance with the defender example is the AI was given the goal of surviving. The next step would be to put the AI in an environment, and let it come up with its own goals and attempt to accomplish them. But what would drive the need to come up with a a goal? How would it determine a "good" goal worth pursuing vs a "bad" goal that should be rejected?
I do think "flashes of brilliance" is running some sort sort of algorithm (whether a genetic algorithm, or something else) in our subconscious. How many times have you been stuck on a problem, then go to bed, and when you wake up, the solution hits you? Something was going on while you slept.
In many ways we are all just highly advanced pattern matching machines. That sudden sense of dread you get when you walk into some scenario is likely your subconscious pattern matching something in your environment with a past negative experience.
One nuance with the defender example is the AI was given the goal of surviving. The next step would be to put the AI in an environment, and let it come up with its own goals and attempt to accomplish them. But what would drive the need to come up with a a goal? How would it determine a "good" goal worth pursuing vs a "bad" goal that should be rejected?
I do think "flashes of brilliance" is running some sort sort of algorithm (whether a genetic algorithm, or something else) in our subconscious. How many times have you been stuck on a problem, then go to bed, and when you wake up, the solution hits you? Something was going on while you slept.
In many ways we are all just highly advanced pattern matching machines. That sudden sense of dread you get when you walk into some scenario is likely your subconscious pattern matching something in your environment with a past negative experience.
This is what I've been thinking about AI as well, they're just pattern machines. I don't like how some people's brains turn off when "AI" is associated with it. It's still only an advanced script. The difference is that a script starts from zero and then learns, improves and evolves. AI is given access to pools of "knowledge" which it can use as part of its learning process and improvements. This allows it to advance faster because it doesn't have to start from zero.
It's like the difference between wise and smart. (Learning from someone else's mistakes vs learning from your own). Currently assuming the knowledge pool that AI is drawing from is considered accurate, AI is considered both wise and smart. Advanced scripts with no outside data access is only smart.
But all it takes is faulty pools of knowledge and the AI is basically neither wise nor smart. Just like humans with higher level intelligence (ie: MBAs, PHDs etc.), they are still capable of GIGO type of outputs. (ie: 130 IQ person spouting inaccurate stereotypes like it's truth).
TIL, CHATGPT might be a potential equalizer in the field of spousal arguments.
"If you bake 75 more cookies than cakes for a total of 575, how many cookies are there?"
"325"
"WTF, are you sure?"
"Don't ask me if I'm sure, ask ChatGPT how to do it"
This will go viral very very quickly. Try it out now while it's still free. Near Sora level for free (30 generations a month). I fully expect the free generations to get limited quickly. Insane that on the same day we get the highly anticipated Stable Diffusion 3 Medium release (which is a colossal fail that will likely cause Stability AI's end) we get this.
I used one of my Midjourney images I had created for a video test I was planning out for a faceless Youtube channel with the prompt I built for the story, and it's providing a very coherant smooth video on just the 2nd attempt. The closest video tools available before this were PikaLabs, Runway, Haiper and SVD. They were limited to 2 to 4 seconds and FPS was abysmal.
I mean it's good, like really, really good. There is nothing else remotely close to it available to the public right now. Luma emerging out of nowhere like this is insane.
All i need is to polish it up a bit in DaVinci Resolve, and run scenes based off a storyboard, add music from Suno, voiceover and sound effects from Elevenlabs, and you have a full AI generated video that could be polished enough to pass the eye test.
With LunaLabs Dream Machine I think we do have semi-commercially viable video capabilities.
The Following User Says Thank You to Firebot For This Useful Post:
Will people even have spouses anymore with all these sexbots in the near future?
A buddy of mine was wondering the other day if for financial and emotional health reasons polygamy and polyamory will begin to skyrocket. Like a controlled open relationship with additional relationship, household management and other benefits.
With LunaLabs Dream Machine I think we do have semi-commercially viable video capabilities.
I don't know, everything I've seen from it still looks pretty hokey. Even the best examples are still pretty 2D looking, but they rely on lots of zooms and pans to dress it up and hide the flaws.
And I don't think it's even clear that the current path of using LLM-based generative AI to produce video will ever even be successful in commercial-level production, except maybe for some stock footage where the details don't matter too much.
If they can figure out a way to take properly done 3D models and then animate them based on text input, then we'd definitely have something that could be used for high-level production. But my understanding is that's estimated to be pretty difficult to achieve at this point.
The Following 2 Users Say Thank You to opendoor For This Useful Post:
That is all you need to know about the general population's ability to discern reality. The threshold needed is not as high as is being made here.
Sora (which is still considerably better than this model) has been estimated to be a 8/10 by commercial video producers with 10/10 being Hollywood level. I would say this is a 5/10 or 6/10 from my use of it and seeing examples. I never stated this is high-level production grade level yet, just that it passes the eye test.
Consider that Luma was released less than 24 hours ago and this was hastily created and people are already making this level of quality. Just refine or re-run the hokey sections that don't pass the eye test (the first scene being an obvious one). Also consider how long such a short film would take to animate, you will absolutely see companies use this in real use applications. We just have to look at CGI's evolution and use in films to see how easily AI can be used. I mean look at this thing...that is from a 98 million budget. And it probably cost 6 figures to do alone. It made the cut. Look at Cats in 2019 and the original Sonic and both are recent.
It seems every time something new comes out in the AI world there is nitpicking that occurs because it's not perfect. I agree that this might not be the right path forward and may be limited. Yet you blink and something better comes along that becomes the new level. Look back at what we had a year ago, 2 years ago. You will have bad AI and good AI use. Just good enough for commercial production is really the threshold for use and I see this as already being achieved in video which is remarkable.
On a side note I estimate I am at 2500$ USD in mostly passive revenue from AI image generations via subscription and commissions since I started this particular side stream a year ago. This includes commissions for custom fantasy book characters and T-shirt logos on AI models.
I don't know, everything I've seen from it still looks pretty hokey. Even the best examples are still pretty 2D looking, but they rely on lots of zooms and pans to dress it up and hide the flaws.
And I don't think it's even clear that the current path of using LLM-based generative AI to produce video will ever even be successful in commercial-level production, except maybe for some stock footage where the details don't matter too much.
If they can figure out a way to take properly done 3D models and then animate them based on text input, then we'd definitely have something that could be used for high-level production. But my understanding is that's estimated to be pretty difficult to achieve at this point.
This is a very common line of thinking but it feels like many aren't extrapolating where it's going or how fast it's getting there. What's giving people shoulder-shrugs today would have been cast off as pure witchcraft a year and a half ago. Will it level out? Maybe, but I'm not sure we're seeing evidence of that yet (I could be wrong).
I don't think what you're saying about commercial applications sounds off at all, but I do think that the history of tech is positively swimming in examples of times we said "I don't think <thing> will ever get powerful enough to..." followed shortly by it happening. There's always some unexpected chaos that gets added to the mix that people don't see coming.
I've played around with Luma, and while it's not going fool everybody, it's spooky good, especially for a company that seemingly just sorta showed up.
There is also no shortage in tech of things that initially seem amazing, but getting them beyond that is proving to be very difficult. See self driving vehicles...
The Following User Says Thank You to Fuzz For This Useful Post:
There is also no shortage in tech of things that initially seem amazing, but getting them beyond that is proving to be very difficult. See self driving vehicles...
Self driving vehicles are an excellent example, but that's also an incredibly difficult problem to solve that–conceivably if this AI trajectory continues–we'll solve. It also doesn't help that the main player behind self-driving vehicles is something of a maniac that overpromises almost constantly and is horny to destroy himself.
Self-driving cars aside, I do feel there's more on the other side of the equation. Oddly enough, cars themselves fall into this category, as does the internet and all the computers it runs on.
Nobody can tell the future, but when we have to bet on a guess, I'm thinking continuous advancement is typically the safest bet.
Good bet for it to keep advancing, but will it hit the same sort of practical wall, where it's often pretty great, but you can't trust it, or it never gets fingers right or whatever. Musk said self driving was easy. Many companies went all in, like Uber, thinking they could get there. No one has yet , a decade later.
. And it probably cost 6 figures to do alone. It made the cut. Look at Cats in 2019 and the original Sonic and both are recent.
