Re: [agi] Re: putting models in your robots head

2019-10-30 Thread geoffreykublin
All is needed Architecturally wise the layers themselves can be arbitrary but for most sequence processing people will shout "Transformer!" and call it a day Most processes do follow a sorta-autoregressive thing and sequence modeling is a good way to look at the problem, so I get the jump to th

Re: [agi] Re: putting models in your robots head

2019-10-30 Thread geoffreykublin
Learning is important question and I do not want to take away from it, but these can be treated independently -- Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T8fe5317c3cebf70b-M683fd2c37cb525d42124bb21 Delivery op

Re: [agi] Re: putting models in your robots head

2019-10-30 Thread immortal . discoveries
https://ai.stackexchange.com/questions/10623/what-is-self-supervised-learning-in-machine-learning Self supervised learning is also when you talk to yourself, you squeeze out and mine gold from what you know, so it's supervised. but not by us lol. Itself! Unsupervised! Anyway, it's learning, ok?

Re: [agi] Re: putting models in your robots head

2019-10-30 Thread doddy
self supervised learning is another one. is self supervised learning better than one shot imitation learning? what about self supervised imitation learning? is there a such thing? On Wed, Oct 30, 2019 at 7:04 PM wrote: > The issue for me is different generalized classes of tasks operate on > dif

Re: [agi] Re: I completely hate today's mainstream AI (Google etc.)

2019-10-30 Thread immortal . discoveries
The only people advancing AGI seems more and more to be larger companies like Open AI and Google. They get more mentions. They are super rich now and famous and have a huge team. They re-use their architectures across OpenAI5, the Arm, GPT-2, MuseNet, Image Completer, Hide & Seek. And they give

Re: [agi] Re: putting models in your robots head

2019-10-30 Thread geoffreykublin
The issue for me is different generalized classes of tasks operate on different sets of symbols, and we often use AI that are trained to operate on a single set of symbols, often the symbols are dependent on the AI as well, as with GPT-2 token embedding and position embedding, or any Transformer

Re: [agi] Re: I completely hate today's mainstream AI (Google etc.)

2019-10-30 Thread Steve Richfield
The AI problem is analogous to the UFO problem - there appears to be something there that works, but no one has a clue as to even where to start to figure out how it works. Steve On Wed, Oct 30, 2019, 2:58 AM wrote: > oh! damn I hate misinterpreting things, I need to concentrate harder when > i

Re: [agi] Re: putting models in your robots head

2019-10-30 Thread geoffreykublin
Transformers make cool hive minds, narrow AI obviously -- Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T8fe5317c3cebf70b-M563c158dd2da494a07f0aa66 Delivery options: https://agi.topicbox.com/groups/agi/subscription

Re: [agi] Re: putting models in your robots head

2019-10-30 Thread keghnfeem
 Me as the worlds top AGI scientist, GPT-2  is not AGI. It is still narrow AI.  AGI is made up of many  narrow AI's to work in the same body. But some of these narrow in the bunch generate other narrow AI's.  In a completer unsupervised manner.   I do not see these special AI's generating and t

Re: [agi] Re: putting models in your robots head

2019-10-30 Thread immortal . discoveries
unsupervidsed  one-shot  imitation-learning  inter-agent  self-play Did yous ever think GPT-2 could self-play itself through imitation? Imitation keeps the topic same but also learns knowledge quickly by passing it down, which requires communicating with other related agents. ---

Re: [agi] Re: putting models in your robots head

2019-10-30 Thread keghnfeem
  I see unsupervised learning is still very primitive, in the world of machine learning. -- Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T8fe5317c3cebf70b-Mb05f719a17457fbb2fafc0e6 Delivery options: https://

Re: [agi] Re: putting models in your robots head

2019-10-30 Thread doddy
here is link with a definition for one shot learning. https://paperswithcode.com/task/one-shot-learning one shot imitation learning. Our approach is to combine meta-*learning* with *imitation learning* to enable *one*-*shot imitation learning*. The core idea is that provided *a single* demonstrati

Re: [agi] Re: putting models in your robots head

2019-10-30 Thread rouncer81
please explain? -- Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T8fe5317c3cebf70b-M20982232dbbd1a96d3535b44 Delivery options: https://agi.topicbox.com/groups/agi/subscription

Re: [agi] Re: putting models in your robots head

2019-10-30 Thread doddy
you could look up one shot learning. it is interesting. you could look up the definition of one shot imitation learning. On Wed, Oct 30, 2019 at 8:49 AM wrote: > Im not sure what that stuffs about - but if you can imitate something, you > can guess what itll do next. > *Artificial General Intel

[agi] Re: AI Learns To Compute Game Physics In Microseconds ⚛️

2019-10-30 Thread rouncer81
Thats cool, its definitely paralleling mine too.. especially since you can run physics engine on the gpu. (hawhaw) Yes, a physics engine cant do it alone, because theres some things you cant form geometry for,  sometimes it too quick for the camera to store,  sometimes its hidden from the camera

[agi] Re: AI Learns To Compute Game Physics In Microseconds ⚛️

2019-10-30 Thread keghnfeem
 I think it is great idea that parallels my work.  There are some flaws for doing it in a AGI  fashion. I am not say what they are.      -- Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/Ta6b978def84fbe9f-M3a5da5763

[agi] Re: AI Learns To Compute Game Physics In Microseconds ⚛️

2019-10-30 Thread rouncer81
Physics engine optimization.    Sounds like a good idea!  Because when your building a motor network, you have to sample the physics for all its possible mutations, so there is alot to reiterate. What do you think about it? -- Artificial General Intelligenc

[agi] AI Learns To Compute Game Physics In Microseconds ⚛️

2019-10-30 Thread keghnfeem
AI Learns To Compute Game Physics In Microsecond: https://www.youtube.com/watch?v=atcKO15YVD8 -- Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/Ta6b978def84fbe9f-M374aa13ff326b58d14466756 Delivery options: https:

Re: [agi] Re: putting models in your robots head

2019-10-30 Thread rouncer81
Im not sure what that stuffs about - but if you can imitate something, you can guess what itll do next. -- Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T8fe5317c3cebf70b-M698c3eae44877dfac6491f1e Delivery options:

Re: [agi] Re: putting models in your robots head

2019-10-30 Thread doddy
I like looking one shot imitation learning videos. have you looked at this website. https://bair.berkeley.edu/blog/ On Wed, Oct 30, 2019 at 5:23 AM wrote: > that is cool Doddy. > > 3d computer vision is the way for me too, because it removes the single > perspective from the detection. The fi

Re: [agi] Re: putting models in your robots head

2019-10-30 Thread rouncer81
that is cool Doddy. 3d computer vision is the way for me too, because it removes the single perspective from the detection.   The first computer vision was 2d, and it still works, but 3d is the cheeky alternative that should work just as well with less stuffing around with neural network mumbo

Re: [agi] Re: putting models in your robots head

2019-10-30 Thread doddy
facebook is doing it too. here is a link. https://venturebeat.com/2019/10/29/facebook-highlights-ai-that-converts-2d-objects-into-3d-shapes/ On Tue, Oct 29, 2019 at 2:44 PM doddy wrote: > you posted the video on youtube today. > > On Tue, Oct 29, 2019 at 12:42 PM wrote: > >> had that music blas

Re: [agi] Re: I completely hate today's mainstream AI (Google etc.)

2019-10-30 Thread rouncer81
oh! damn I hate misinterpreting things, I need to concentrate harder when im reading other ppls posts...  god sorry. Yeh! haha! watch out its a small world!!! -- Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T10119

Re: [agi] Re: I completely hate today's mainstream AI (Google etc.)

2019-10-30 Thread Stefan Reich via AGI
Nooo, I just meant that my publicly declared hate is now a thing one can find online... On Wed, 30 Oct 2019 at 10:48, wrote: > Why is that Stefan - ask me more questions and ill help clarify things if > u want me to. > Im not sure if these things I say are to be said, or kept secret... > becaus

Re: [agi] Re: I completely hate today's mainstream AI (Google etc.)

2019-10-30 Thread rouncer81
Why is that Stefan -  ask me more questions and ill help clarify things if u want me to. Im not sure if these things I say are to be said, or kept secret... because people arent supposed to know? Here is my unit - and its using search, and it seems quite lively. https://www.youtube.com/watch?v=6H

Re: [agi] Re: I completely hate today's mainstream AI (Google etc.)

2019-10-30 Thread Stefan Reich via AGI
Sigh. Now this mail's subject is haunting me... On Wed, 30 Oct 2019 at 09:56, wrote: > I wouldnt call it a complete ripoff - a search can kick your butt at > chess, or any other domain for that matter, and it actually can seem > "alive" if a computer looks all ends and picks the max. The fut

Re: [agi] Re: I completely hate today's mainstream AI (Google etc.)

2019-10-30 Thread rouncer81
I wouldnt call it a complete ripoff - a search can kick your butt at chess, or any other domain for that matter, and it actually can seem "alive" if a computer looks all ends and picks the max.     The future of a.i. is searching, I dont think its even moral to make something that can think like

Re: [agi] Re: Missing Data

2019-10-30 Thread rouncer81
Yes thats true in a way I.D.,   but u have to supply a parser and calculator with the decompressor for it to work that out, and it only works on basic arithmetic, nothing else, just given that example. -- Artificial General Intelligence List: AGI Permalink: