Re: [agi] Re: OpenAI just announced Sora, and it's incredible

[immortal . discoveries]

Ah, haha, darn it, I see, you went from transistors slowing out and yes other 
thingy exists - the slow right-now molecular computing you mentioned, which is 
far off currently. Though like I said I guess, I hope AI is its own thingy 
here, to boost us up like moore's law. It seems to grown large now and not done 
growing, and almost there to help stop our ageing.

Once we get human AI, while it seemed to me intelligence and their robotics 
movements will double and take off, it's possible both will stall and that 6 or 
20 years of progress they give us a year might not be enough still to do fix 
our bodies ageing.

The one thing that sticks out though is Cryonics seems to almost work now, and 
our arteries should be able to be easily cleaned with a bit updated tech, and 
cancer seems like something we are closing in on solving. And lastly, what 
really sticks out is one way that seems to be a fast all we need moore's law 
would be the cloning nanobots eating the soil and land and basically the world, 
if it worked out we would seem to have a larger amount of what we need very 
fast, as the growth would occur fast.

I can visualize, I can see the amount of data and atoms in a cell, in our human 
volumewe are not at that level of computing our codes, to understand how to 
tinker with us to save and shape us, yet, for a while. It seems like though if 
you got the nanobots to clone and be useful, this threshold would be surpassed 
Very Fast. Because if you have trillions of zillions of nanobots, that would be 
a lot more computing and other things like a lot more AI laws etc growing fast, 
faster than our current factories make "computing"/ transistors.

So, hopefully nanobot cloning will bring us to the compute power we need, fast, 
to work with the human sized systems and not just molecules like we are at 
basically stuck at right now . Simple molecules is where we are at I mean to be 
clear.
--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T4ad5d8c386d0e116-M56af19a1f2a7afb63f2df0ea
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Re: [agi] Re: OpenAI just announced Sora, and it's incredible

[Matt Mahoney]

The doubling time for Moore's law, 1.5 years, is just one of many
technologies that underwent exponential growth for awhile, just like the
automotive industry from 1890 to 1920. If it weren't for the limits of
physics, we should have cars today that travel faster than light and cost
less than a dollar. We saw similar predictions for nuclear power and space
travel in the 1950s. Now transistors are at the limits of physics, and
other technologies like genomics are growing exponentially, like all
innovations do in the early stages.

The overall effect of many different technologies is centuries of economic
growth (relative to the price of food) at 2-3% per year, or a doubling time
of 25-30 years. Any trend on a shorter time scale won't make reliable
predictions beyond that scale.

For now we can look forward to life expectancy increasing at 20% (0.2 years
per year), declining population in developed countries as more people live
alone, more government control while feeling more free, more economic
inequality even as the poor get richer, open borders, abolition of arrests
and prisons (but not crime), and the disappearance of cash and non
synthetic meat. We are at the transition of technology connecting us to
isolating us as we interact more with AI and less with other people. Large
events, stadiums, theaters, and shopping centers will disappear, along with
all forms of mass entertainment, shows and music, in favor of private
virtual worlds with private languages in our smart homes served by self
driving delivery carts.

Beyond the next century, the most persistent trend is the chaotic nature of
evolution, punctuated by mass extinctions and rapid replacement with new
species, events that form the boundaries of geologic eras. We are in one
now, very close to replacing DNA based life with something more efficient.

--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T4ad5d8c386d0e116-M458bdea2d94ca3500c26b422
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Re: [agi] Re: OpenAI just announced Sora, and it's incredible

[immortal . discoveries]

Aren't you assuming Moore's Law is the only thing happening there to move 
progress ahead? You said doubling every 1.5 years...  But what happens when we 
have 8 billion humanoid robots, for example. Now they move and think 10x faster 
in just a couple of years next. Then 20x faster, for example. Now we have not 
doubling every 1.5 years, but doubling 20x faster than "1.5" years.
--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T4ad5d8c386d0e116-M498daecb12a00c1d4490108e
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Re: [agi] Re: OpenAI just announced Sora, and it's incredible

[Matt Mahoney]

Here are Manifold's predictions for SORA by the end of 2025. There is a 12%
chance they can produce a 20-40 minute video. There is a 77% chance OpenAI
will be sued, 30% chance of being banned in at least one EU country. Video
will almost certainly be watermarked.
https://manifold.markets/Bayesian/what-will-be-true-of-openais-sora-m

I am being generous with Moore's law. It is actually slowing down. Clock
speeds stalled in 2010. Transistors are near their physical size limit of a
few nanometers, the spacing between N or P dopant atoms in silicon. The
only way to reduce power consumption significantly is to compute by moving
atoms or molecules instead of electrons. A human brain sized neural network
using transistors needs a few thousand GPUs and 1 MW of electricity. The
brain uses 20 watts because neurons compute by moving sodium and potassium
ions that are 100,000 times slower and heavier than electrons but use
100,000 times less energy. DNA bases and amino acids are on the order of
another 100 times slower, heavier, and energy efficient than neurons.

Before the earth is overrun by nanobots, we have to transition from
transistors to molecular computing. That technology is a long way off. Just
using Moore's law to estimate how long, it will take about a century to
match the 10^37 bits stored in all the DNA in the biosphere.

I know we will get there. Wheels are faster than legs. Jets are faster than
birds. The atmospheric carbon cycle is 210 billion tons per year (20% of
the biosphere) implying a cycle time of 5 years and a power consumption of
500 TW of solar power (at 4 Kcal/g of carbohydrate) out of 90,000 TW of
available sunlight. Global electricity production is only 18 TW but solar
cells are already 20-30% efficient. Electric motors are over 90% efficient,
compared with 22% for muscle.

Near term, advances in AI will come from optimizations like hardware for
sparse, low precision vector operations and specialized vision models that
include a fovea and eye movements. Video will track your eye movements and
leave out details where you are not looking.

This will happen in parallel as molecular computing is developed to replace
transistors. But that is a long way off.



On Fri, Feb 23, 2024, 10:12 AM  wrote:

> On Wednesday, February 21, 2024, at 7:05 PM, Matt Mahoney wrote:
>
> I was not impressed with the music clips, but that's just me.
>
> This is shocking because on my end I have made a buffet of impressive and
> diverse masterpieces. And I can tell this is close to perfect they only
> didn't give us a longer generation limit lol! And took away the AI from us
> too. Even the mario soundtrack? It came prefect. Was desert themed and
> snake sounding etc. The ice world castle ones were perfect. Make sure to
> listen to them all lol.
>
>
>
> On Wednesday, February 21, 2024, at 7:05 PM, Matt Mahoney wrote:
>
> I'm estimating 30 years assuming that the cost of generating a movie or
> song using AI drops by half every 1.5 years
>
> half.only half every 1.5 years :) . But you wouldn't say this
> after the planet is swarming with ASIs and nanobots, I don't think... I
> think it works like that maybe for another 2 years and then we really get a
> accelerated progress.
>
> AGI going to finally surpass us, humans have been stuck forever at human
> level. And human speed.
>
> Even just humanoid robots, at human speed, can run around and move their
> fingers really fast, in human-speed bounds. We don't really run around and
> move fast because it costs us more energy and seems like more work, but
> that is our downfall. A human can actually run move 4 times faster than
> typically do. That would be 4 times the progress per year. They will also
> think 4 or 100 times faster. And more intelligent. This is an explosion of
> progress coming it seems.
>
> Yes, to make it clear, I can over my kitchen table move my fingers and
> hands 6 times faster than normally would be seen, i.e. I can move my finger
> (like I am pointing at X or pressing a button) in/out from a fist staring
> position like 10 times per secondyou get it. It looks like motion blur
> its so fast.
>
> In fact I can't think fast enough to use that human-speed maximum of my
> motors. So it's mostly useless to humans therefore. But the robots will
> have much more time to think about their movements.
>
>
>
> But anyway, GPT-5 is coming this year probably and will be AGI. ASI next
> year :) Brett said too robotics is solves now he scaling up and making
> incredible progress we will prob see it make fences etc this year and do
> everything! If it can do what we saw making a coffee for the most part, it
> is all done, we are so close.
>
> Video AI is solves seeing Sora. Music AI is solved. Text AI solved once
> reach GPT-4. Image AI solves with DALL-E 3 nearly (got my hard test to
> flunk it but they were not expecting my hard tests haha!).
> *Artificial General Intelligence List *
> / AGI / see discussions 

Re: [agi] Re: OpenAI just announced Sora, and it's incredible

[immortal . discoveries]

On Wednesday, February 21, 2024, at 7:05 PM, Matt Mahoney wrote:
> I was not impressed with the music clips, but that's just me.
This is shocking because on my end I have made a buffet of impressive and 
diverse masterpieces. And I can tell this is close to perfect they only didn't 
give us a longer generation limit lol! And took away the AI from us too. Even 
the mario soundtrack? It came prefect. Was desert themed and snake sounding 
etc. The ice world castle ones were perfect. Make sure to listen to them all 
lol.



On Wednesday, February 21, 2024, at 7:05 PM, Matt Mahoney wrote:
> I'm estimating 30 years assuming that the cost of generating a movie or song 
> using AI drops by half every 1.5 years
half.only half every 1.5 years :) . But you wouldn't say this after the 
planet is swarming with ASIs and nanobots, I don't think... I think it works 
like that maybe for another 2 years and then we really get a accelerated 
progress.

AGI going to finally surpass us, humans have been stuck forever at human level. 
And human speed.

Even just humanoid robots, at human speed, can run around and move their 
fingers really fast, in human-speed bounds. We don't really run around and move 
fast because it costs us more energy and seems like more work, but that is our 
downfall. A human can actually run move 4 times faster than typically do. That 
would be 4 times the progress per year. They will also think 4 or 100 times 
faster. And more intelligent. This is an explosion of progress coming it seems.

Yes, to make it clear, I can over my kitchen table move my fingers and hands 6 
times faster than normally would be seen, i.e. I can move my finger (like I am 
pointing at X or pressing a button) in/out from a fist staring position like 10 
times per secondyou get it. It looks like motion blur its so fast.

In fact I can't think fast enough to use that human-speed maximum of my motors. 
So it's mostly useless to humans therefore. But the robots will have much more 
time to think about their movements.



But anyway, GPT-5 is coming this year probably and will be AGI. ASI next year 
:) Brett said too robotics is solves now he scaling up and making incredible 
progress we will prob see it make fences etc this year and do everything! If it 
can do what we saw making a coffee for the most part, it is all done, we are so 
close.

Video AI is solves seeing Sora. Music AI is solved. Text AI solved once reach 
GPT-4. Image AI solves with DALL-E 3 nearly (got my hard test to flunk it but 
they were not expecting my hard tests haha!).
--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T4ad5d8c386d0e116-M14370bc99132dd7c3f13a624
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Re: [agi] Re: OpenAI just announced Sora, and it's incredible

[Matt Mahoney]

>From your examples it is obvious that text to image and video is rapidly
improving. No doubt we will see this technology in movies by next year. But
there is still a big step to custom generated music replacing mass media in
about 30 years. Video game graphics have vastly improved over the last 40
years to the point we don't need human actors any more, but we still need
humans to direct and curate the movie, knowing what an audience will like,
even if that just means experimenting with prompts.

I was not impressed with the music clips, but that's just me. Artists
already have some advanced tools, like keyboards that can sound like any
instrument. Streaming services can recommend songs based on what you
listened to, liked, or skipped in the past. We already have apps that can
identify songs by artist and title just by listening. But unlike video, we
don't have a language for describing music much beyond that.

I'm estimating 30 years assuming that the cost of generating a movie or
song using AI drops by half every 1.5 years, when the average audience size
drops from 1 million to 1. I think music generation is a harder problem
than video. With video an AI can learn your tastes by tracking eye
movements and facial expressions, but music mostly lacks these clues.

On Sat, Feb 17, 2024, 11:31 AM  wrote:

> *First see my last reply above, I showed something before but maybe you
> missed it I guess.*
>
>
>
> On Friday, February 16, 2024, at 1:16 PM, Matt Mahoney wrote:
>
> It costs $1M to $10M to produce a movie for a mass audience that you would
> pay $1 to $10 to watch. If Moore's law drops the price by half every 18
> months, then we are still 25 years from customized movie genres replacing
> mass entertainment. Should we expect a similar time scale for music?
>
>
> Below is 3 AIs, each a year apart, don't you see what's happening? Ya,
> movie generation is going to be all done by 2025.
>
> 2021 Nov 24:
> https://github.com/microsoft/NUWA/blob/main/NUWA.md
>
> 2022 Oct 5
> https://imagen.research.google/video/
>
> 2 days ago:
> https://openai.com/sora
> *Artificial General Intelligence List *
> / AGI / see discussions  +
> participants  +
> delivery options 
> Permalink
> 
>

--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T4ad5d8c386d0e116-M0cf8f52ed0acbaf2f1fc7509
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Re: [agi] Re: OpenAI just announced Sora, and it's incredible

[immortal . discoveries]

*First see my last reply above, I showed something before but maybe you missed 
it I guess.*



On Friday, February 16, 2024, at 1:16 PM, Matt Mahoney wrote:
> It costs $1M to $10M to produce a movie for a mass audience that you would 
> pay $1 to $10 to watch. If Moore's law drops the price by half every 18 
> months, then we are still 25 years from customized movie genres replacing 
> mass entertainment. Should we expect a similar time scale for music?
> 

Below is 3 AIs, each a year apart, don't you see what's happening? Ya, movie 
generation is going to be all done by 2025.

2021 Nov 24:
https://github.com/microsoft/NUWA/blob/main/NUWA.md

2022 Oct 5
https://imagen.research.google/video/

2 days ago:
https://openai.com/sora
--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T4ad5d8c386d0e116-M099ae48c25a18052ea57c3fe
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Re: [agi] Re: OpenAI just announced Sora, and it's incredible

[immortal . discoveries]

On Friday, February 16, 2024, at 1:16 PM, Matt Mahoney wrote:
> Is audio a harder problem than video, or does video have a bigger payoff so 
> they attacked it first?
> 
https://www.reddit.com/r/singularity/comments/13h0zyy/i_really_crank_out_music_tracks_with_musiclm_this/
--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T4ad5d8c386d0e116-M8174e5db9b511ffa559c0e6a
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Re: [agi] Re: OpenAI just announced Sora, and it's incredible

[Shashank Yadav]

Video clearly has a lot more commercial and military applications. Moreover, 
audio alone doesn't really offer a simulation of physical world with object 
permanence etc. Now there won't be user customized movies right away, but the 
present production process surely is going to get upended, and not to mention 
the areas like advertising and porn. The capability increment here wrt humans 
is significant, so perhaps we're not very far from the exponential slide.     





regards, 

https://muskdeer.blogspot.com/ 







 On Fri, 16 Feb 2024 23:46:38 +0530 Matt Mahoney  
wrote ---







On Fri, Feb 16, 2024, 1:33 AM   wrote:



https://openai.com/research/video-generation-models-as-world-simulators






So many questions. How much training data, how much compute to train, how much 
compute to generate a video, how many parameters?



It is estimated (because OpenAI didn't say) that GPT4 has 1.8 trillion 
parameters, was trained on 13 billion tokens (most of the public Internet), 
cost $100M to train (10^25 ops at 10^17 per dollar) and runs on 128 A100 GPUs 
(around $2M). 
https://the-decoder.com/gpt-4-architecture-datasets-costs-and-more-leaked/

But other sources give different guesses.



Is audio a harder problem than video, or does video have a bigger payoff so 
they attacked it first?



It costs $1M to $10M to produce a movie for a mass audience that you would pay 
$1 to $10 to watch. If Moore's law drops the price by half every 18 months, 
then we are still 25 years from customized movie genres replacing mass 
entertainment. Should we expect a similar time scale for music?








https://agi.topicbox.com/latest / AGI / see https://agi.topicbox.com/groups/agi 
+ https://agi.topicbox.com/groups/agi/members + 
https://agi.topicbox.com/groups/agi/subscription 
https://agi.topicbox.com/groups/agi/T4ad5d8c386d0e116-Mc9f3c07a5bb0dac06360e952
--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T4ad5d8c386d0e116-M3dec15e64cdb652d277cade3
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Re: [agi] Re: OpenAI just announced Sora, and it's incredible

[Matt Mahoney]

On Fri, Feb 16, 2024, 1:33 AM  wrote:

> https://openai.com/research/video-generation-models-as-world-simulators
>

So many questions. How much training data, how much compute to train, how
much compute to generate a video, how many parameters?

It is estimated (because OpenAI didn't say) that GPT4 has 1.8 trillion
parameters, was trained on 13 billion tokens (most of the public Internet),
cost $100M to train (10^25 ops at 10^17 per dollar) and runs on 128 A100
GPUs (around $2M).
https://the-decoder.com/gpt-4-architecture-datasets-costs-and-more-leaked/
But other sources give different guesses.

Is audio a harder problem than video, or does video have a bigger payoff so
they attacked it first?

It costs $1M to $10M to produce a movie for a mass audience that you would
pay $1 to $10 to watch. If Moore's law drops the price by half every 18
months, then we are still 25 years from customized movie genres replacing
mass entertainment. Should we expect a similar time scale for music?

--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T4ad5d8c386d0e116-Mc9f3c07a5bb0dac06360e952
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Re: [agi] Re: OpenAI just announced Sora, and it's incredible

[James Bowery]

OpenAI has lobotomized DALL-E to the point that it is virtually worthless for 
continuity.  I doubt they've done much better for switching between scenes of 
motion pictures.  You can't do such simple things as storyboards or graphic 
novels.  The way they achieved this feat of engineering?

By closing off the only way of getting a face to be the same recognizable 
character between frames -- which is to specify it as a well-sampled celebrity. 
They yammer at you how this is some sort of violation.  So what do you do to 
get it to render a face consistently between frames?  You describe the 
character in detail and hope that it will do something reasonable.  Here, for 
example, is what I was limited to doing for a storyboard I wanted to do:

https://twitter.com/jabowery/status/1755378830426915008 


https://twitter.com/jabowery/status/1755382127510564904


--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T4ad5d8c386d0e116-Mc34b9d32018972272eb7c796
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Re: [agi] Re: OpenAI just announced Sora, and it's incredible

[mm ee]

Based on how they describe it, they're using patching and represent videos
as grids of diffused images. Definitely some clever attention work going on
too, since occluded objects also remain pretty stable and also between
camera cuts. The model size is speculated to be not that high either (rumor
is 3-5 billion parameters, definitely runnable on a consumer GPU.) The real
magic seems to be in the synthetic training data and their captioning
recipe, which IMO, so many labs are behind them when it comes to labeling
quality (If anyone remembers, the big reason GPT-3 and ChatGPT 3.5 were so
good at the time was the sheer volume of high quality labeled data they got
from ScaleAI. It's a solid formula that hasn't really been tapped outside
their org)

On Fri, Feb 16, 2024, 8:30 AM stefan.reich.maker.of.eye via AGI <
agi@agi.topicbox.com> wrote:

> I'm officially speechless. Check out the waves example at
> https://openai.com/sora. This is photorealism. Are they now simulating
> physics too? I wouldn't know how you could distinguish this from a real
> video. How did they advance the field by this much with a single release?
> OpenAI has shown their superior inspiration once again.
> *Artificial General Intelligence List *
> / AGI / see discussions  +
> participants  +
> delivery options 
> Permalink
> 
>

--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T4ad5d8c386d0e116-Mfeb74075e425525d673e35a7
Delivery options: https://agi.topicbox.com/groups/agi/subscription

[agi] Re: OpenAI just announced Sora, and it's incredible

[stefan.reich.maker.of.eye via AGI]

I'm officially speechless. Check out the waves example at 
https://openai.com/sora. This is photorealism. Are they now simulating physics 
too? I wouldn't know how you could distinguish this from a real video. How did 
they advance the field by this much with a single release? OpenAI has shown 
their superior inspiration once again. 
--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T4ad5d8c386d0e116-M7b407449a6ad8ed7805334b0
Delivery options: https://agi.topicbox.com/groups/agi/subscription

[agi] Re: OpenAI just announced Sora, and it's incredible

[stefan.reich.maker.of.eye via AGI]

On Friday, February 16, 2024, at 12:23 PM, immortal.discoveries wrote:
> https://twitter.com/sama/status/1758219575882301608
So the spoon materializes out of thin air, moves strangely and then dissolves 
into the bowl. Other than that, I didn't see a single artifact. This is nuts.

> sam posted some new ones to his twitter 

So you're on a first name basis now, are you? LOL
--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T4ad5d8c386d0e116-M2b446cba4f7929bdde4a17ad
Delivery options: https://agi.topicbox.com/groups/agi/subscription

[agi] Re: OpenAI just announced Sora, and it's incredible

[immortal . discoveries]

sam posted some new ones to his twitter damn so good!
https://twitter.com/sama/status/1758219575882301608

go to his account from this link okl, there's a few!
--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T4ad5d8c386d0e116-Me41ad1fa64fb1cb716cd5b94
Delivery options: https://agi.topicbox.com/groups/agi/subscription

[agi] Re: OpenAI just announced Sora, and it's incredible

[immortal . discoveries]

https://openai.com/research/video-generation-models-as-world-simulators
--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T4ad5d8c386d0e116-Mefbc68f2c524c51935538c45
Delivery options: https://agi.topicbox.com/groups/agi/subscription