Full transcript
0:00Things like app development will
0:01disappear , too . If I were 22 , maybe I
0:03wouldn't go to Microsoft now .
0:05Oh , it changed . So my opinion has
0:07changed .
0:08Well , rather than that , it's better to
0:09choose based on your own sense of
0:11conviction .
0:11Yeah .
0:12That's probably hard for job seekers ,
0:14isn't it ? It seems tough to know what
0:17to be convinced by .
0:18But , well , I only do work that I’m
0:20convinced by . It’s like a
0:21notification pops up saying , " Oh ,
0:23something strange just happened . "
0:25Yes . Being able to properly catch up
0:27when that " something strange " happens
0:29is very important ; getting bored is
0:31actually quite significant . Oh , can't
0:33you just job hunt until you get bored
0:35of it ? So ...
0:36Job hunt until you get bored .
0:38If you do that , you'll reach a point of
0:39conviction . Once you realize , " I'm
0:42bored now , " if you have a core , you'll
0:44be fine .
0:45I think everyone is struggling with how
0:46to build that core right now . Like 22 -
0:48year-olds .
0:49Well , it's true that AI can do anything
0:50now .
0:51Exactly .
0:51How should we build it ? Try building it
0:54and see .
1:06Hello .
1:07Hello .
1:08Hello .
1:09Hello .
1:11Hello .
1:14I don't need the picture-in-picture .
1:16I'll add it now .
1:17Yes .
1:21Please . Once again ,
1:24Once again , what kind of program is
1:26this ?
1:26This time , it's about " If I were 22
1:29today , " 22
1:30years old , 16 years ago , 16
1:33years ago . Mr. Ochiai .
1:34Yes .
1:35Uh , in 2017 ...
1:37Yeah .
1:37I interviewed you on NewsPicks ,
1:39and at that time , you said that if you
1:41were 22 , you’d want to go to
1:43Microsoft .
1:43Ah , that was around 2017 , maybe nine
1:46years ago ?
1:47That's right .
1:48Ah , Microsoft just after releasing the
1:50HoloLens . Right , right .
1:52Ah ,
1:52How is it ? Has your thinking changed
1:54from back then ?
1:55Well , Microsoft is still a good company
1:57, isn't it ? I'm still on good terms
1:58with them . Yeah .
1:59The world I saw then and the world I
2:01see now are basically the same — or
2:03rather , the world I was imagining .
2:05I mean , AI is there , hardware is strong
2:07, and the platformers are still strong
2:09as ever , so I don't really feel like
2:11much has changed .
2:12That's true .
2:12Yeah .
2:13I feel like I wrote my book " Survival
2:15in the Super AI Era " back in 2017 .
2:17That's right .
2:18Yeah .
2:182017 . It said that what humans can do
2:21is muscles .
2:21That's true , isn't it ?
2:22Yes . Mostly correct .
2:24Correct . I feel like you're saying the
2:25same thing now .
2:26Oh , muscles .
2:27Muscles .
2:28Also , you didn't recommend IT companies
2:30close to production houses .
2:32Ah .
2:32Like production work .
2:34What about that ?
2:35Because app development and such will
2:36disappear .
2:37Yeah . It'll be automated .
2:38It'll disappear , right ? It already has .
2:41Seriously .
2:42It's gone .
2:42I mean , even web development is almost
2:45gone , isn't it ? Increasingly .
2:47Because it's getting automated . In that
2:49case , people making a living on
2:51platforms are safe , but the platforms
2:53themselves will be built by AI .
2:55Platforms too .
2:56Yeah . So that's open to debate , but ...
2:59Yeah .
3:00I feel like the gap is closing rapidly
3:02now .
3:02Certainly . We can do it ourselves now ,
3:04too .
3:05Yeah . Quite a lot .
3:06Yeah .
3:07Yeah .
3:07By the way , in an interview nine years
3:09ago , you said that companies like
3:10Toyota might be good , surprisingly .
3:12Yeah . I don't think automobiles are bad
3:14at all , even now . I mean , they're
3:16working hard , saying autonomous driving
3:18is coming soon . With AI emerging , I
3:20feel like companies with hardware might
3:22be able to survive .
3:24Yeah .
3:26In other words , companies with
3:28production facilities are higher up the
3:30supply chain than startups that just
3:32build hardware .
3:34After all , hardware is becoming
3:36incredibly important right now , isn't
3:38it ? Yeah .
3:40Maybe it's about balance , or
3:41challenging both if possible ?
3:44Well , infrastructure for intelligence
3:46is being allocated now , so ...
3:51Things like being smart or solving
3:53problems with technical software
3:54resources have mostly dropped in cost .
3:58But I feel like whether or not you have
4:00strong hardware is still crucial .
4:02Yeah .
4:02Yeah .
4:03I see . Whether it's " safe " or not , are
4:05there any industries or genres that
4:07seem interesting ?
4:08Interesting industries ? Right now , well
4:11, let's see . But there is a possibility
4:13that a " humanoid winter " might be
4:15coming . A winter is coming .
4:16I feel like you can't just lump them
4:18all under the term " humanoid , " though .
4:21I mean , it would be great if current
4:23systems could solve most humanoid
4:25problems , but in terms of action and
4:27reaction , the problems they are solving
4:29right now aren't really suited for
4:31lifting heavy objects .
4:33Yeah .
4:33Well , they might be able to dance , but
4:35they're still not very good at carrying
4:37things .
4:38There are quite a few issues , and I
4:39think it's possible that many demo
4:41applications were created , but they
4:43just didn't fit into real-world
4:44operations .
4:45Huh . Yeah . It might drop for a while
4:49and then rise again , but while everyone
4:51is caught up in the frenzy , we haven't
4:53fully tested what they are truly good
4:55or bad at in the field .
4:58Huh . There’s talk about domestic
5:00robots arriving by 2027 , isn't there ?
5:02Yes .
5:02Will they really arrive ?
5:04Well , lots of them will be released ,
5:05right ?
5:06Then , after the release , it'll be a
5:07matter of whether people actually use
5:08them . I mean , Roomba essentially
5:10collapsed .
5:10That’s true .
5:11Yeah . It's a world where even iRobot
5:13can go under . So , I don't think there's
5:15any guarantee that a company is safe
5:17just because it's a humanoid company .
5:19Basically , if it doesn't meet the needs
5:21of the market , people won't use it .
5:23Yeah .
5:24Yeah . I think the turnover is quite
5:26fast .
5:27Like , physical AI isn't going as well
5:28as we thought ?
5:29Well , there are all sorts of physical
5:31AI out there , so I feel like it's
5:33generally going well , but —
5:35Yes .
5:36I don't think there are that many
5:37places where humanoids can work better
5:39than humans .
5:40Yeah .
5:41Yeah . I mean , sure , that might be the
5:43case in the future , but it's pretty
5:44hard to predict exactly when things
5:46will really take off .
5:47Yeah . I see . I think OpenAI has brought
5:50about amazing innovation . Should we
5:53look at leaders like Sam Altman to
5:55determine if a company is promising , or
5:57how else should we evaluate their
5:59future potential ?
6:01But even the creators of ChatGPT didn't
6:03think ChatGPT was going to be useful .
6:06Oh .
6:06Yeah . I mean , the text prediction model
6:09came out first anyway .
6:11Yeah . Yeah . Yeah .
6:13But people were surprised at how
6:14addictive the conversational one was
6:16because it was so easy to use . It was
6:18addictive , and users would even correct
6:20its mistakes , which was great for
6:22reinforcement learning and gathering
6:24datasets .
6:26That's true . I watch NVIDIA . Really ?
6:30Oh , ever since I was in grad school ,
6:32when people asked me where to invest ,
6:34I’d tell them to bet on NVIDIA .
6:37Since back then ?
6:38Yes .
6:39Wow .
6:40Amazing . How did you know , or rather ,
6:42why ?
6:43Because deep learning semiconductors
6:45only really worked with NVIDIA , and
6:47rewriting CUDA programs was a huge pain
6:49.
6:50Yes .
6:51So , I had a feeling it would just keep
6:53growing .
6:54Wow . Was that the main point of
6:57interest while you were doing research ?
7:00Was it because you found it useful , or
7:02perhaps because it was versatile ?
7:04Something like that .
7:05No , honestly , NVIDIA was just all about
7:08GPUs back then , so in my slides at the
7:10time ,
7:11I wrote , " Why is deep learning this big
7:13? Thanks to all the gamers out there ,
7:15you're the best . " Yeah .
7:16For people who want to become engineers
7:18now ,
7:18Yeah .
7:19Since everything is automated , are
7:21there fewer opportunities to learn ?
7:23Yeah . True . Well , if you've done it
7:25from scratch , you can do anything ,
7:27right ? It’s like you develop a gut
7:29feeling for it . Yeah .
7:31Having that intuition is pretty
7:32important for an engineer ; if you can't
7:35build tools that are actually useful to
7:37you , it's hard to become an engineer in
7:39the first place .
7:41And beyond just being useful to
7:42yourself , it’s hard to build things
7:44that others will find useful unless you
7:46really think it through .
7:48Yeah .
7:49So , yeah . That might be true . You know ,
7:52in the US , there’s talk of
7:54engineering jobs disappearing , while in
7:56Japan , it’s still ...
7:58Maybe a bit safer , I suppose . At 22 ,
8:00you're looking for a job and it's a
8:02situation where you can probably still
8:04find one , but ...
8:05Well , for Japanese engineers , of course
8:07there are a lot of software engineers ,
8:10but engineering in Japan is often
8:11focused on cars or parts , which is
8:14super niche , or rather , the kind of
8:16engineering that people in mechanical
8:18or materials engineering do ,
8:20I mean , sure .
8:21Solving equations or predicting things
8:23with software can be handled quite well
8:24by AI , but ,
8:25well , you still need expertise , don't
8:27you ? That's engineering for you .
8:28Yeah , I guess so . Ultimately , for
8:30someone listening to this who's 22 ,
8:32companies like Toyota seem like a good
8:34place to be .
8:34Well , wouldn't that be fine ?
8:36Right . From a job-hunting perspective .
8:38Yeah , I think it's fine . But I'm not
8:40sure which way things will go .
8:41Oh .
8:42Well , it's hard to say if platformers
8:44will be able to maintain their status .
8:46I see . So , if you could go back to
8:49being 22 , would you still want to go to
8:52Microsoft ?
8:52If I were 22 right now .
8:54Right , right now .
8:55If I were 22 today , well , being 22 . 22 ,
8:59yeah . If I were 22 , I might not go to
9:01Microsoft now .
9:02Oh , you've changed your mind . Then ...