Re: [agi] True AI Technology

[Matt Mahoney]

Would it work? I can't tell because your plan is very vague. Meanwhile, the
big tech companies are already at the realization phase with language
models that pass the Turing test. The path to AGI now looks like more
powerful hardware to implement vision and robotics.

On Fri, Oct 13, 2023, 10:09 AM  wrote:

>
> *technology*
>
> As humanity globally approaches a true AI system, it becomes increasingly
> clear that there will not be one true AI system. Instead, there will come
> into existence many different true AI instances with various
> characteristics. According to this expectations, in my project, I assemble
> the process of true AI creation from building three distinctive layers:
> *foundation*, *plan*, and *realization*.
>
> *foundation *(...under construction...)
>
> Foundation represents a medium for defining plans. The foundation should
> be abstract enough to inspire planner's creativity, but also expressive
> enough not to limit provided possibilities. I'm currently in the phase of
> building such a foundation.
>
> *plan *(...among interests...)
>
> Plan is a program coded in foundation terms, representing a species from
> which an AI being may be instantiated. Obviously, a number of different
> possible species can be created, each carefully envisioned and planned by
> the creators of true AI.
>
> *realization *(...far in the future...)
>
> Realization is a plan in execution, representing an operating AI being
> that learns from its environment, applying learned knowledge to its
> behavior. Collected personal knowledge would be the most important
> distinguishing point between different AI beings instantiated from the same
> species.
>
> ---
>
> What do you think, would this work? More or less unintentionally, it
> somehow aligns with the Nature.
> *Artificial General Intelligence List *
> / AGI / see discussions  +
> participants  +
> delivery options 
> Permalink
> 
>

--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/Td02eb9a7e06e7b5e-Mf77e5fd9fc4f5a6b13ffa85c
Delivery options: https://agi.topicbox.com/groups/agi/subscription

[agi] Re: True AI Technology

[immortal . discoveries]

I may be missing something to grasp this, I'm not sure, but, I'm not sure what 
any of this means, or what caused you to feel the need to make it, etc.  How 
does it help create AGI if AGI is what perhaps OpenAI is building? There is a 
lot of work they put into it, along with others, I think that needs the most 
attention, including a plan to safely run the AGI once made. But I'm not sure 
what your writing above is about, or how it helps with that. Also there is a 
way to evaluate for AGI, using hard and easy tests, though I also don't see you 
mention that here either.

Perhaps too rigid and not natural enough? Correct me if I am very wrong but 
don't you guys like the older way of AI I think so this is kinda giving me the 
GOFAI vibes.

Foundation plan realization sounds like memory, planning, and acting. I don't 
see anything new here then I guess?? That's what brains do I thought.
--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/Td02eb9a7e06e7b5e-Mdbc5d9fa29b14680b8663dd4
Delivery options: https://agi.topicbox.com/groups/agi/subscription

[agi] True AI Technology

[ivan . moony]

*technology
*

As humanity globally approaches a true AI system, it becomes increasingly clear 
that there will not be one true AI system. Instead, there will come into 
existence many different true AI instances with various characteristics. 
According to this expectations, in my project, I assemble the process of true 
AI creation from building three distinctive layers: *foundation*, *plan*, and 
*realization*.

*foundation *(...under construction...)

Foundation represents a medium for defining plans. The foundation should be 
abstract enough to inspire planner's creativity, but also expressive enough not 
to limit provided possibilities. I'm currently in the phase of building such a 
foundation.

*plan *(...among interests...)

Plan is a program coded in foundation terms, representing a species from which 
an AI being may be instantiated. Obviously, a number of different possible 
species can be created, each carefully envisioned and planned by the creators 
of true AI.

*realization *(...far in the future...)

Realization is a plan in execution, representing an operating AI being that 
learns from its environment, applying learned knowledge to its behavior. 
Collected personal knowledge would be the most important distinguishing point 
between different AI beings instantiated from the same species.

---

What do you think, would this work? More or less unintentionally, it somehow 
aligns with the Nature.
--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/Td02eb9a7e06e7b5e-M41dd6046eb07704c162a6d02
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Re: [agi] GPT Vision is unbelievable

[immortal . discoveries]

Yes. There is also a rumor OpenAI has AGI now also. GPT-4 was made a year ago, 
they withheld it for a few months. There is also rumor Google employees are now 
scared of OpenAI too.

It is named Arrakis.
https://medium.com/@daniellefranca96/rumors-about-openai-having-a-model-more-powerful-then-gpt4-bce4f2520603

Also one OpenAI employee on Twitter said WAGMI - We're all gonna make it.

I tried this GPT-4 vision, it was on Bing for free for the last maybe 1 or 2 
months I have it noted when pretty exactly in my folders if need it. New test 
by someone in paid version now, finally called GPT-4, and, same failure though 
on captcha test for saying which images have crosswalks. Yes, it works 
sometimes, even gets them all! But it fails a ton too on it. Same for paid 
version called politically GPT-4. It is free on Bing, same for DALL-E 3, in 
fact that says DALL-E 3 powers it, it's free, and doesn't change ur prompt by 
GPT-4 saying hey I'll change it toand doesn't say NO to yoshi, etc 
characters lol, and you pay too, eh! Sad, wtf. Free version is better! So then 
this means I tried GPT-4 vision then ya...

I told yous :D, GPT-4 is Weak AGI. AGI comes 1 or 2 years after that. It is not 
just me saying such things, yes it is known by others this such timelines and 
speed of how this works. Though I am tracking it and have a good bird's eye 
view of how true this might be.

Text AI is faster and ahead of image AI. Video AI even farther behind. Same for 
AI with audio, too. AGI will likely wake up as GPT-5 with mostly just text 
ability. Mostly. Take a look at how fast DALL-E bombed other AIs and its older 
selves, look below! This speed is for IMAGE AI, clearly it will be perfect in 2 
years. Text AI, 1 year. But that's public use, it is probably makeable now, in 
lab, expensive, but can make 1 GPT-5, and test it, slowly.
https://www.reddit.com/r/dalle2/comments/176lch7/dalle_3_dalle_25_dalle_2_mj50_mj_4_stability_ai/

Much of my GPT-4 tests are below. I showed a robot experiment in my own home. 
You can see my kitchen hehe :). Pretty smart! I should have asked it to make 
small 10 degree turns at a time yes. Still interesting at that part of the 
video though!
https://www.reddit.com/r/bing/comments/15m53h4/huge_upload_of_all_my_one_is_robot_test_hard/

Yous also seen from my other recent posts MusicLM tests I showed, are stunning, 
it's close also to human level, aside from lack of audio abilities and such of 
course.
https://www.reddit.com/r/singularity/comments/13h0zyy/i_really_crank_out_music_tracks_with_musiclm_this/

Robots are coming by the dozens, and also AI with them, able to do almost 
anything now. AIG is around the corner.

Driving video AI, wow, check out the new updated one! It's dreaming, and gonna 
give us perhaps self driving cars now, especially if coupled with GPT-4! There 
is another research where it talks to itself why it is turning and ex. the 
video shows the car speaking saying from dashboard view I mean in captions "i 
am now slowing down due to a protruding car, now speeding up since clear" etc.
Here it is below I mean:
They had an update below. Both though can generate very very long coherent 
videos, which is cool. They showed in the other prior update blog post that it 
can generate 4 alternative samples of what video comes next, it can generalize 
and make alternative videos, it is powerful, that's so great!
https://wayve.ai/thinking/scaling-gaia-1/

I actually have a hard time showing GPT-4 is NOT human, using not math but 
story completion testing. I prefer story testing because it is the whole thing 
about the context, it is what it seen most, it is like a movie, if it can do it 
human than that is what I want to see, I don't even need test it on math or 
cost... My tests though for text to image AI are REALLY ADVANCED and perfected 
now haha, ahead of time, and indeed my hardest stump even DALL-E 3. But DALL-E 
4 might get all them but my toughest 1 or 2. I think it might take DALL-E 5 
though to ace all them lol. That's then 1.5 or 2.5 years then away using 
current ! trend speed. BTW DALL-E can even copy cartoon and make what I want it 
to be, look! Not all them came perfect but this did, aside from wrogn girl 
being there lolz, 
https://www.reddit.com/r/dalle2/comments/172ux8h/vicky_from_fairy_odd_parents_is_standing_in_the/

I'm going at GPT again though to find a tougher test, besically at the END of 
human ability guys. Here is it started already it already fails GPT-3, and is 
though not showing really yet failure in GPT-4 though really, but will soon 
once done. Look at that guys, GPT-3 fails here!:

gpt4 try1
https://justpaste.it/d8tfx

gpt4 try2
https://justpaste.it/bv5f5
---
gpt3 try1
https://justpaste.it/c45pv

gpt3 try2
https://justpaste.it/9c4zh

gpt3 try2
https://justpaste.it/c38hn

Gotta love how GPT-3 though adds to its output at end "It is the oddest thing 
ever and very unnecessary wordage in this story