Re: It's all about power...
lets see what QUALCOMM does.. and the OSS community, im usign ChatGpt wieslawsoltes on github to make small agents or ask chatgpt to code them from scratch with minimal UI using Avalonia, so i can drive any OS from on AI. hopefully. at least for dictation and some navigation. a dynamic mic helps. i cant use al these different UIs anymore ill bite the bullet and make my own. hopeful if my calls and puts bets work i can get help..
Meta , OpenAi , and MSFT fund the ways to get around the cuda moat.
https://medium.com/@1kg/nvidias-cuda-monopoly-6446f4ef7375
mfst reported traing took down a city grid, and they had to distribute it.
Low suply, high demand say from Elon musk who's ai is barely used,Grok?
bought ridiculous amounts of NVIDIA H100 chips for each at 15 to 25k .
We could have a moon base for just a sliver of the Trillions given to NVIDIA market cap , and train AIs on solar abundant power on the Lunar south pole wiht superconducting ambient conditions for free, theoretically. its depressing. a herd giving a trillion dollars when the our species is the last apex predator, ecosystem at a tipping point, and our technology still does not have an off planet backup, and the moon is 2days away!!
that article above was before everyone wants an ARM chip , laptops replaced towers, but my laptop is too hot to touch.
Apple doesn't have a touch screen they want you to have a ipad and a macbook air so Surfaces and 20 others have a green field.. im temped to get one jut because its summer and i dont' have air-conditioning.
So developer facing, and faster with batter life 6 -16 hours not 1 hours, is enough to immediately change, imo.. with people workin mostly offsite now.
they have this huge opportunity. electricity and isn't getting cheaper and it boils the planet.. Latency, hotword training, DSPs being programmed, reversible fixed point solvers, and such, i want much of my AI local and custom, i cant use copilot, wont. So if more developers /researchers have this QUALCOMM DSP array and FPGA and NLP and not use the graphics card for anything but graphics, and
chat gpt helps me implement OSS based custom chatgents each with a prompt, .. I used Voice Assess and then ChatGPt to fix it. instead of RAG..but ill try to learn to run it locally. im mostly want low latency on the voice to text and OCS to drive my pc hands free.
this is QUALCOMMS pitch but its a clusterfk
https://www.slideshare.net/slideshow/accelerating-newer-ml-models-using-the-qualcomm-ai-stack-a-presentation-from-qualcomm/258398945
https://cloud.google.com/blog/products/ai-machine-learning/vertex-ai-nas-makes-the-most--advanced-ml-modeling-possible..
NVIDIA employees are also 3% of the holders. they are very rich and lazy now.