0:00
a bunch of people at anthropic write code.
0:02
So we made this internal tool called Cloud Code,
0:05
and because we ourselves write code,
0:07
we have I think a special and unique insight
0:10
into how to best use the AI models to write code.
0:14
So I think in the code space,
0:17
we become very strong competitors.
0:21
>> What industry do you think will get disrupted
0:25
and what has a certain runway left?
0:27
>> I'm asking from the lens of,
0:30
I'm trying to figure out what book to read,
0:33
which college to go to, what skillset to learn.
0:38
If I'm starting a startup today,
0:42
what has some kind of a tailwind?
0:45
>> Yeah.
0:45
>> For a short period of time is okay as well.
0:47
>> Yeah, I mean, I would think about tasks that are human-centered,
0:53
tasks that involve relating to people.
0:57
I think that the stuff like code and software engineering
1:00
is becoming more and more kind of AI-focused,
1:04
things like math and science.
1:06
>> Is that coding or engineering?
1:07
If I were to segregate coding and engineering
1:10
to be two completely different things?
1:12
>> Yeah.
1:12
>> Is coding going away or is engineering element of software
1:16
where you're an architect trying to figure out?
1:19
>> I think coding is going away first,
1:21
or coding is being done by the AI models first.
1:24
And then the broader task of software engineering
1:26
will take longer.
1:28
But I think that doing that end to end,
1:31
I think that is going to happen as well, I would say.
1:34
But again, the elements of design
1:38
or making something that's useful to users
1:40
or knowing what the demand is,
1:41
or managing teams of AI models,
1:45
those things may still be present.
1:48
Again, this comparative advantage is surprisingly powerful.
1:52
Even if you're only doing like 5% of the task,
1:57
like that 5% gets super amplified and levered,
2:00
because it's like you're only doing 5% of the task,
2:03
the AI does the other 95%,
2:04
and so you become 20 times more productive.
2:08
Again, at some point, you get to 99% and then it becomes harder.
2:11
But I think there's surprisingly much in that,
2:15
in that sort of, in that zone of comparative advantage.
2:19
But I would really think about the things
2:21
that are human-centered.
2:22
I think there's something to that.
2:26
I think there's something to kind of the physical world,
2:29
or things that mix together,
2:31
human-centered, the physical world,
2:35
one of those two, and analytical skills
2:37
that somehow tie them together,
2:39
similar to the radiologist example I gave.
2:42
- So what would I study?
2:43
Say, I'm actual use case, I'm 25 years old.
2:47
I'm trying to pick a profession for myself.
2:49
I want some kind of tailwind.
2:51
My outcome is a capitalistic win in the next decade.
2:55
What industry would I pick outside of something
2:59
which has a physical interface?
3:00
- Yeah, again, anything where you're building on AI,
3:03
like if AI is the tailwind,
3:05
if you can be part of some other part of the supply chain,
3:08
something in the semiconductor space,
3:10
which I think is, that's one example.
3:15
It has an element of kind of physical world
3:18
and more traditional engineering, not software engineering.
3:22
Again, the very kind of human-centered professions,
3:25
like that is something I would think in terms of.
3:29
And I think the other thing I always say
3:30
is like in the world in which AI can kind of generate anything
3:34
and create anything, having basic critical thinking skills
3:39
may be the most important thing to success.
3:41
I worry about these AI models that generate images
3:45
and videos, and we don't make models
3:47
that generate images and videos for many reasons,
3:50
but this is one of them.
3:51
It's really hard to tell what's real from what's not.
3:56
And so a significant part of success
3:59
may be having the street smarts,
4:01
not to get fooled by it.
4:03
I mean, hopefully we can crack down on
4:06
and regulate some of this fake content,
4:09
but assume we can't, critical thinking skills
4:14
are gonna be really important.
4:15
And you don't wanna fall for things that are fake.
4:18
You don't wanna have false beliefs,
4:20
you don't wanna get scammed.
4:22
That's really advice that I would give to someone.
4:25
- If every innovation in the history of humanity
4:30
killed a core human skills, I'll give you an example.
4:33
If calculators killed our ability to do arithmetic,
4:37
if writing reduced the memory of human beings per se,
4:43
what muscle is AI killing?
4:45
- So, first of all, I'm not so sure.
4:49
I still do math in my head quite a lot.
4:53
I still find it useful to do math in my head,
4:57
even without a calculator,
4:58
just because it's more integrated
5:01
into my thought processes, right?
5:03
I might wanna say, oh yeah,
5:05
it's like each user paid this amount,
5:07
then the revenue would be that.
5:09
I wanna be able to close that loop in my head
5:11
without having to give the answer to a calculator.
5:15
So, I think a lot of these skills are still pretty relevant.
5:18
But, I would say that if you don't use things carefully,
5:23
that you can lose important skills.
5:26
And I think we started to see it with students
5:31
where they have the AI, write the essay for it.
5:35
It's basically just cheating on homework.
5:36
So, we shouldn't do that.
5:38
We did some studies around code
5:40
and showed that depending on how you use the model,
5:44
we can see de-skilling in terms of writing code, right?
5:47
There are different ways to use the model,
5:49
and some of them don't cause de-skilling,
5:51
and some of them do.
5:51
But, definitely, if folks are not thoughtful
5:54
in how they use things, then de-skilling absolutely can happen.
5:58
- Do you think humans will become stupid as a race
6:02
in the next decade?
6:03
'Cause if we are, in a way,
6:05
exporting, thinking and cognition
6:09
do systems? - Yeah, I think if we deploy,
6:13
again, it's the machines of loving grace
6:15
and adolescence of technology.
6:17
I think if we deploy AI in the wrong way,
6:20
if we deploy it carelessly, then, yes,
6:23
people could become stupid.
6:24
Even if an AI is always gonna be better than you at something,
6:27
you can still learn that thing, right?
6:29
You can still enrich yourself intellectually.
6:32
And so, that's a choice we have to make
6:34
as individual companies, as individual people,
6:36
and as a society overall.
6:38
When I tried to use Cloud Code for the first time,
6:42
I did struggle to get it to work.
6:45
It was for somebody who's very stupid
6:48
and has no coding or programming knowledge,
6:50
it's not very, very easy.
6:53
I think there's a learning curve.
6:55
I heard someone say it well,
6:57
it's like even prompt engineering is like playing a piano.
7:01
You can't sit and start playing it.
7:03
To my audience, I think it becomes increasingly relevant
7:07
where to learn how to set context, how to prompt,
7:12
how to use Cloud Code better for somebody like me
7:15
who comes with zero knowledge.
7:17
Can you recommend how one does that?
7:19
- Yeah, I mean, first of all, I would say,
7:21
we're trying increasingly to make that learning curve easier.
7:26
So, one of the things that caused us to release
7:28
Cloud Co-work, which is basically a Cloud Code
7:31
for non-coders is, oh man, we were noticing
7:35
a bunch of non-technical people
7:37
who really wanted to use Cloud Code
7:39
and were struggling through the command line terminal
7:41
to do that.
7:42
It's like coders use the command line terminal all the time,
7:45
but non-coders just kind of makes things
7:49
unnecessarily complicated.
7:51
So, co-work was designed to be more of the kind of,
7:56
it was powered by the Cloud Code engine on the back,
8:00
but the idea was to kind of make it more like user friendly
8:05
and easier to use.
8:07
So, we're definitely trying to introduce interfaces
8:11
that kind of make it easier,
8:12
but I would also say that there's like,
8:17
classes you can take that help you learn this thing.
8:20
Now, I think it's a very empirical science
8:22
you mostly learn by doing,
8:23
but it's like anthropic has it's like,
8:26
part of the company that we call the Ministry of Education,
8:29
and I think increasingly we'll put out videos
8:32
on how to run effective agents and how to prompt models.
8:35
We've already done some of that,
8:36
and I think we're gonna ramp it up,
8:39
'cause we do want everyone to be able to learn this.
Reply