Andy Bechtolsheim: AI consumes most of the compute cycles at Google
From the Goldman Sachs Tech Conference 2018, Andy Bechtolsheim (co-founder of Sun and the first investor in Google, prior to the company even being founded):
"AI consumes most of the compute cycles at Google, and it is growing.
I have been interested in AI for a long time. My first programming project was a neural network simulator in 1974, on a PDP-11 with 64kB of memory and we could only model 256 nodes.
And today, AI is the most exciting place in all of high tech. In normal IT, you might see 10-20% improvement per year. Here you see an order of magnitude improvement each year.
Analog AI chips would be the lowest power. The human brain is remarkably power efficient, and it is analog. [I agree. Go Mythic!]
NVIDIA has more software engineers than hardware engineers.
So much in life can be improved by 10%. Think about sales, CRM, insurance. Every vertical has something that can be improved with AI.
NVIDIA's Huang: There is now 10x as much computer-generated code as human-generated code. [e.g., when you train a neural net; 90% of code is computer generated].
Andy Bechtolsheim: AI consumes most of the compute cycles at Google
From the Goldman Sachs Tech Conference 2018, Andy Bechtolsheim (co-founder of Sun and the first investor in Google, prior to the company even being founded):
"AI consumes most of the compute cycles at Google, and it is growing.
I have been interested in AI for a long time. My first programming project was a neural network simulator in 1974, on a PDP-11 with 64kB of memory and we could only model 256 nodes.
And today, AI is the most exciting place in all of high tech. In normal IT, you might see 10-20% improvement per year. Here you see an order of magnitude improvement each year.
Analog AI chips would be the lowest power. The human brain is remarkably power efficient, and it is analog. [I agree. Go Mythic!]
NVIDIA has more software engineers than hardware engineers.
So much in life can be improved by 10%. Think about sales, CRM, insurance. Every vertical has something that can be improved with AI.
NVIDIA's Huang: There is now 10x as much computer-generated code as human-generated code. [e.g., when you train a neural net; 90% of code is computer generated].