So they have billions to spend. Will they waste it like they did on OpenAI which just developed bad search engines and video games. Ok to be kind GPT-3 is a moderately OK but not great search engine but hugely wasteful, not green due to thousands of hours of electricity to train it on big parallel computers and in the end smaller more intelligent engines like Noonean's EnterpriseNLP outperform it.
They also wasted billions developing "responsible AI" with ZERO concept of developing cognitive systems and language. Gee real smart. I guess as long as Ilya Sutskever and the Hinton-ists from Google Brain (another failed research group accomplishing almost NOTHING) are calling the shots and getting huge 2M a year salaries, they will bullshit their way to years of money accomplishing little.
Yes, they had groundbreaking Image Recognition. No that does not apply to developing cognitive systems.
These systems always develop trainable finite state analysis systems because that is the limit of tensor models. Same with deep mind playing GO. Games are finite state -> response modeling. Easy. Cognition and language - hard. You cannot develop cognitive systems (at least not good ones) using 2D tensor multi-planar architectures. Yet all the libraries for GPUs force you into a tensor mindset. Chopping your legs off from the start. Where is JAVAs GPU library to develop real architectures? Not yet sadly.
Noonean has been working on Neural Cubes (technically octagons) which are fully dynamic running at 100Hz or 100 updates a second. That is fast enough to respond to the world in real time. The numbers of neurons to build a small system to do this is roughly 8 billion. Each requiring 500 operations per cycle. So roughly 4 TFLOPs just to get started of FP16 (FP32 generally is not needed). The differences? It is self organizing self connection and self structurally building whereas TENSOR models are like flat rooms all constructed ONCE and never changing. It is the difference between a Boeing 777 and a pile of dirt.
What about IBM Watson efforts? Again it suffers from poor architecture making limited use of neural networks and more use of their knowledge expression language. Again forcing it into rigid systems. Still powerful things can be done with the IBM approach and it is superior to the wasted billions being done at Google Brain and OpenAI.
Facebook recently announced an initiative to do multiple integrations into a system for cybernetics (robots). they don't have any new technologies but it is a strong move forward. Someone tell Facebook that Noonean could help them a lot! Maybe they will develop clunky death robots like Boston Dynamics (now google) Atlas? Yes Atlas can do a backflip. Sometimes. More often it steps into a hole in the ground and falls on its face.
These systems are heavy and not nimble. They do balance, but they don't have balance. their fundamental design is flawed. Can you get stepper motors and servos to slightly replicate human notion? Barely. Reminds me of the horrific ECCO bot which used servos turning strings as "more human muscle" ya right. Bullshit on that.
It should be like the manhattan project, but without a cognition system, not simply a barrier recognition system like in self driving cars, these platforms will always seem... well.. dumb. That's not the goal. The goal is to have a assistant that humans would accept into their work and homes and smart enough to get by, handle black swan events with aplomb and if there is another fukushima or space emergency we can send in a troop of them to save the day. That's the vision Noonean has had since 2018 and has been working towards.
Will Tesla's efforts result in another cybertruck window crashing. Oh well at least they can make rockets.
