00:00:00
this is a case where necessity was the
00:00:01
mother of
00:00:03
invention these guys were like well how
00:00:05
am I going to do this reinforcement
00:00:06
learning thing they invented a totally
00:00:08
different algorithm it uses a lot less
00:00:10
computer memory and it's highly
00:00:11
performant and then the second thing
00:00:13
that was crazy is everybody is used to
00:00:15
building models and compiling through
00:00:17
Cuda which is nvidia's proprietary
00:00:19
language and these guys worked totally
00:00:21
around Cuda and they did something
00:00:22
called PTX which goes right to the bare
00:00:24
metal we meaning the West with all the
00:00:27
money that we've had didn't come up with
00:00:29
these ideas and I think part of why we
00:00:31
didn't come up is not that we're not
00:00:33
smart enough to do it but we weren't
00:00:34
forced to because the constraints didn't
00:00:36
exist and so I just wonder how we make
00:00:39
sure we learn this principle meaning
00:00:41
when the AI company wakes up and rolls
00:00:43
out of bed and some VC gives them 200
00:00:46
million maybe that's not the right
00:00:48
answer for a series A or a seed and
00:00:50
maybe the right answer is 2 million so
00:00:52
that they do these deep seek like
00:00:54
Innovations